Here is some good information given below;

Good site for information for understanding how to prepare for actuary exams :  http://www.pstat.ucsb.edu/instruction/actuary/study.html

Quizzes and study material can be gotten at

• ASM  - http://www.studymanuals.com/Shop
• ADAPT - https://www.actexmadriver.com/default.aspx

2 popular  Coaching institutes

1. https://www.coachingactuaries.com/ can be used for 30-60 days subscription for practice exams and question bank
2. this below one is more popular http://www.theinfiniteactuary.com/ (also called as TIA) – has good video lectures

 Blog and forums to get practice exams and solutions to problems :

• http://www.actuarialninja.com/actuarial-exams/pass-actuarial-exams/
• http://www.actuarialoutpost.com/
• http://beanactuary.com
•  
  

For Indian actuary exams preparation :  http://www.actuarialscienceonline.in/

 List of exams for CAS designation

 http://www.casact.org/admissions/process/

 https://www.soa.org/education/exam-req/resources/edu-txt-manuals.aspx

 Study methods and tips

http://www.beanactuary.org/exams/what/?fa=study-methods-and-tips

 https://www.soa.org/library/newsletters/the-future-actuary/2007/summer/how-to-prepare.aspx

 List of exams to complete

Preliminary Exams:

o    Exam 1 – Probability (credit received through SOA Exam P)

o    Exam 2 – Financial Mathematics (credit received through SOA Exam FM)

o    Exam 3F – Models for Financial Economics (credit received through SOA Exam MFE)

o    Exam LC – Models for Life Contingencies**

o    Exam ST – Models for Stochastic Processes and Statistics**

o    Exam 4 – Construction and Evaluation of Actuarial Models (credit received through SOA Exam C)
**Exams offered through Spring 2015. However, transition rules allow candidates with credit for either Exam ST or LC to take the exam required to obtain credit for a new Associate Exam, Exam S, by Spring 2016.

Associate Exams:

o    Exam S – Statistics and Probabilistic Models***

o    Exam 5 – Basic Techniques for Ratemaking and Estimating Claim Liabilities

o    Exam 6 – Regulation and Financial Reporting (Nation-Specific)

§  Actuarial Institute of Chinese Taipei

§  Canada

§  United States

***New exam offered beginning in Fall 2015 replacing Exams LC, ST, and VEE-Applied Statistical Methods.

To reach the FCAS designation, which takes another two to three years after becoming an ACAS, candidates must complete all ACAS requirements, in addition to Fellow Exams.

Fellow Exams:

o    Exam 7 – Estimation of Policy Liabilities, Insurance Company Valuation and Enterprise Risk Management

o    Exam 8 – Advanced Ratemaking

o    Exam 9 – Financial Risk and Rate of Return

To achieve the CERA designation, candidates must complete:

·         All requirements for CAS Associateship

·         CAS Exams 7 and 9

·         Exam ST9, Enterprise Risk Management Specialist Technical of the Institute and Faculty of Actuaries (U.K.)

·         Three-day Enterprise Risk Management and Modeling Seminar for CERA Qualification

1. Descriptive Statistics

Descriptive statistics are a collection of statistical tools which are used to quantitatively describe or summarize a collection of data. Descriptive statistics aim to summarize, and as such can be distinguished from inferential statistics, which are more predictive in nature.

2. Population

A population is a selected individual or group representing the full set of members of a certain group of interest.

3. Sample

A sample is a subset drawn from a larger population. If this drawing is accomplished in such a manner that each member of the population has an equitable chance of selection, the result is referred to as a random sample.

4. Parameter

A parameter is a value which is generated from a population. If I had all the data of all humans on Earth and generated the mean age, this value would be a parameter.

5. Statistic

A statistic is a value which is generated from a sample. If I calculated the mean age of a subset of humans on planet Earth (much more feasible), this value would be a statistic. Hence, the discipline of statistics.

6. Generalizability

Generalizability refers tot he ability to draw conclusions about the characteristics of the population as a whole based on the results of data collected from a sample. This is ability is not a given, and depends heavily on the nature of sample collection, sample size, and various other factors.

7. Distribution

A distribution is the arrangement of data by the values of one variable in order, from low to high. This arrangement, and its characteristics such as shape and spread, provide information about the underlying sample.

8. Mean

Mean, along with median and mode, is one of the 3 major measures of central tendency, which collectively evaluate an important and basic aspect of a distribution. The simple arithmetic average of a distribution of variable values (or scores), the mean provides a single, concise numerical summary of a distribution. The mean is also likely the most common statistics encountered in general research. Population mean is denoted μ, while sample mean is denoted x̄.

9. Median

The median is the score of a distribution residing at the 50th percentile, separating the top and bottom 50 percent of scores. The median is useful for both splitting a set of distribution scores in half and helping to identify the skew of a distribution.

10. Mode

The mode is simply the score which appears most frequently in the distribution. Multimodal refers to a distribution with more than one mode; bimodal refers to a distribution with 2 modes.

11. Skew

When there are more scores toward one end of the distribution than the other, this results in skew. When the scores of a distribution are more clustered at the high end, the relatively fewer number of scores on the low end result in a tail, with the scenario being referred to as negative skew. Positive skew is when a distribution shows a tail at its high end.

In general, in a negatively skewed distribution, we would expect the mean to be less than the median, while in a positively skewed distribution, we would expect the mean to be greater than the median.

12. Range

One of the most important measures of dispersion, the range is the difference between the maximum and minimum values of a distribution.

13. Variance

Variance is the statistical average of the dispersion of scores in a distribution. Variance is not often used on its own, but can be a useful calculation on the way to a more descriptive statistical measurement, such as standard deviation.

14. Standard Deviation

The standard deviation of a distribution is the average deviation between individual distribution scores and the distribution's mean. Individually, the standard deviation provides a good measure of how spread out a disquisitions scores are. When considered alongside the mean, these 2 measures provides a good overview of the distribution of scores.

15. Interquartile Range (IQR)

The IQR is the difference between the score delineating the 75th percentile and the 25th percentile, the third and first quartiles, respectively.

1.       SAS

2.       IBM SPSS

3.       RapidMiner

4.       H2O.ai

5.       Microsoft

6.       Weka (machine learning)

by harvard

https://www.experfy.com/training/certifications/data-science-certification

CPCU - Certified Property and Casualty Insurance

Cyber Insurance

Associate in Re-insurance (ARe)

Associate in Risk Management (ARM)

Associate in Insurance in Data Analytics (AIDA)

Associate in Insurance (AINS)

ACAA, FCAA by CAS

Chapter 1 - Introduction to Risk Management

• Uncertainty of outcome - risk involves uncertainty about the type of outcome, the timing of outcome, or both the type and timing of the outcome. Uncertainty can be expressed as probabilities or possibilities. 
• Possibility of a negative outcome - possibility means that an outcome or event may or may not occur. Possibility of loss : are evaluated along four dimensions namely : the loss frequency, the loss severity, the total dollar loss and the timing of loss.
•  
  

Possibility compared with probability - the possibility of something occurring does not indicate the likelihood of it happening. The probability is the likelihood of the outcome or event occurring

To quantify risk, one needs to know the probability of the outcome or event occurring. Unlike possibility, probability is measurable and has a value between zero and one.  

Probability : The likelihood that an outcome or event will occur. Unlike possibility, probability is measurable.

Possibility : means that a risk exists and has simply been identified. It does not quantify risk.

Classification of risk

• Pure and Speculative risk
• Subjective and Objective Risk
• Diversifiable and Non-Diversifiable  Risk
• Quadrants of Risk (hazard, operational, financial and strategic)

These classifications are not mutually exclusive and can be applied to any given risk.

Pure Risk : A chance of loss or no loss, but no chance of gain.

Speculative Risk : A chance of loss, no loss, or gain. Is usually created intentionally

Distinguishing between pure and speculative risks is important because those risks must often be managed differently,

Credit Risk : The risk the customers or other creditors will fail to make promised payments as they come due. Relevant to any organization with accounts receivable such as banks or other financial institutions.

Price Risk : change in the cost of raw materials and other inputs, as well as cost-related changes in the market for completed products and other outputs.

Subjective Risk : The perceived amount of risk based on an individual's or organization opinion.

Objective Risk : The measurable variation of uncertain outcomes based on facts and data.

Subjective Risks can exist where Objective Risk does not.The  close an individual's or organization's subjective interpretation of risk is to the objective risk, the more effective its risk management plan will likely be. Subjectivity is also necessary because facts are often not available to objectively assess risk.

Reason the subjective and objective risk can differ include :

1. Familiarity and Control
2. Consequence or Likelihood
3. Risk Awareness

Diversifiable Risk : A risk that affects only some individuals, businesses, or small groups.

Non-Diversifiable Risk (or fundamental risk): A risk that affects a large segment of society at the same time.

Systemic Risk : are generally non-diversifiable. Example : Is the potential for a major disruption in the function of an entire market or financial system.

Quadrants of Risk : 

• Hazard risks arise from property, liability, or personnel loss exposures and are generally the subject of insurance.
• Examples: Direct Damage to property, Indirect damage to property, Terrorism, Injury to employees, third party liability claims.
• Operational risks fall outside the hazard risk category and arise from people or a failure in processes, systems, or controls, including those involving informatio and technology.
• Examples : Product recall, Discrimination, Embezzlement, Workplace Violence, Kidnap, Turnover, Service Provider Failure, Supplier Business interruption
• Financial risks arise from the effect of market forces on financial assets or liabilities and include market risk, credit risk, liquidity risk, and price risk. It can be diversifiable or non-diversifiable
• Examples : Economic changes, Legislative changes, Commodity Prices, Exchange rates, Debt Rating, Liquidity, Currency Exchange Rates 
• Strategic Risks arise from trends in the economy and society, including changes in the economic, political, and competitive environments, as well as from demographic shifts. These are speculative and diversifiable.
• Examples : Intellectual property, Technology, Competition, Acquisition/Merger, Union Relations, Product Design.  

Hazard and Operational Risks are classified as pure risks, and financial and strategic risks are classified as speculative risks. 

Market Risk : Uncertainty about an investments future value because of potential changes in the market for that type of investment.

Liquidity Risk : The risk that an asset cannot be sold on short notice without incurring a loss.

Focus of quadrants of risk is on the risk source and who traditionally manages it. Organizations define types of risk differently. Some organizations consider legal risks as operational risk, and some may characterize certain hazard risks as operational risk.

There can be overlap among the various categories of risk. Each organization should select categories that align with its objectives and processes.

Enterprise Risk Management

traditional risk management is concerned with an organization's pure risk, primarily hazard risk. The concept of enterprise risk management (ERM) was developed in recent years as a way to manage all of an organization's risks, including operational, financial, and strategic risk. 

ERM : An approach to managing all of an organization's key business risks and opportunities with the intent of maximizing shareholder value. Also known as enterprise-wide risk management.

In ERM terms definition of risk management : Coordinated activities to direct and control an organization with regard to risk. 

In ISO terms definition of risk management : The effect of uncertainty on objectives.

all definitions include the concept of of managing all of an organization's risks and its objectives is a key driver in deciding how to assess and treat risks.

Definition of ERM by various bodies

Risk and Insurance Management Society (RIMS) : A strategic business discipline that supports the achievements of an organization's objective by addressing the full spectrum of its risks and managing the combined impact of those risks as an interrelated risk portfolio.

Casualty and Actuarial Society (CAS) : The discipline by which an organization in an y industry assesses, controls, exploits, finance's and monitors risks from all sources for the purpose of increasing the organization's short and long-term value to its stakeholders.

Committee of Sponsoring Organizations of the Treadway Commission (COSO) : Enterprise risk management is a process, affected by an entity's board of directors, management, and other personnel, applied in a strategy setting and across the enterprise, designed to identify potential events that may affect the entity, and manage risk to be within its risk appetite, to provide reasonable assurance  regarding the achievement of entity objectives.

Pillars of ERM 

Risk Managed separately are not the same as thy are when managed together. Three main theoretical concepts explain how ERM works:

• Interdependency - Assumption that statistical independent if the probability of one event occurring, does not affect the probability of a second event occurring.
• Correlation - Correlation increases risk, while uncorrelated risks can reduce risks to the extent that provide a balance or hedge.
• Portfolio Theory - a portfolio is a combination of risks. The portfolio theory assumes that risk includes both individual risks and their interactions.

In ERM , unlike traditional risk management organization model of having a risk manager and risk department to manage hazard risk, the responsibility of the risk management function is broader and includes all of an organization;s risks, not just hazard risks. Additionally the entire organization at all levels becomes responsible for risk management as the ERM framework encompasses all stakeholders.

Dodd-Frank Act, which became U.S. Law in 2010, requires that certain types of financial companies appoint board risk committees. A board risk committees may consist of the full board, the adult committee, r a dedicated risk committee. In addition, some public companies have formed an executive-level risk committee to assist the board in its risk oversight function. The executive-level committee might be chaired by a chief risk office (CRO), who reports to both the Chief Executive Officer, who reports to both the Chief Executive Officer (CEO) and the board risk committee.

As facilitator, the CRO engages the organization's management in a continuous conversation that establishes risk strategic goals in relationship to the organization's strengths, weakness, opportunities, and threats (SWOT). The stakeholders in the organization include employees, management, the board of directors, and shareholders. External stakeholders include customers, regulators, and the community.

In the fully integrated ERM organization, identifying and managing risk become part of every job description and every project. Successful risk management of strategic objectives becomes a measure on all evaluations.

Implementing ERM

• The risk management professional must have access to data from all organization areas and levels to identify and assess the organization's risks.
• The risk management process to manage those risks must be integrated throughout the organization.
• Risk managers must have authority to make and enforce necessary changes, often against significant resistance.
• Effective communication is essential to a successful ERM program
• An organization with a full integrated ERM program develops a communication matrix that moves information throughout the organization.
• The establishment of valid metrics and the continuous flow of cogent data are a critical aspect tho this communication process. The metrics are carefully woven into reporting structures that engage the entire organization, including both internal and external stakeholders.

Impediments to ERM

• Impediments to ERM is technological deficiency
• And traditional organization culture with its entrenched silos, each of the company functions typically had its own management structure.

LOSS EXPOSURES

Any condition or situation that presents a possibility of loss, whether or not an actual loss occurs.

Every loss exposures has three elements

• An asset exposed to loss
• Cause of loss (also called a peril)
• Financial consequences of the loss

Above elements can be described for the four basic types (categories) of loss exposures in which the insured has a financial ineterest.

• Property Loss Exposures : Tangible and Intangible Assets like patents, copyrights
• Liability Loss Exposures : At worst insured/entity may lose total wealth
• Personnel Loss Exposures : Key employee skills that cannot be easily replaced - due to death, disability, retirement or resignation
• Net Income Loss Exposures : Future stream of net income cash flows: Financial Consequences : increase in expenses and loss in revenue/profit

An asset exposed to loss - include property, investments, money that is owed to them, automobiles and cash. Intangible assets includes patents, copyrights and trademarks and human resources for organizations. Individuals may have intangible assets such as professional qualifications, a unique skill set, or valuable experience.

Cause of Loss : 

• Hazard : A condition that increases that frequency or severity of a loss.
• Moral Hazard : A condition that increases the likelihood that a person will intentionally cause or exaggerate a loss. Example: intentionally causing, fabricating, or exaggerating a loss.
• Morale Hazard (attitudinal hazard) : A condition of carelessness or indifference that increases the frequency or severity of loss.
• Physical Hazard : A tangible characteristic of property, persons, or operations that tends to increase the frequency or severity of loss.
• Legal Hazard : A condition of the legal environment that increases loss frequency or severity.

Moral and Morale Hazards are behavior problems that can increases the frequency and/or severity of losses. The fundamental difference between these two types of hazard is intent. A moral hazard results from a deliberate act; a morale hazard results from carelessness or indifference.

Financial Consequences of Loss : depends on the type of loss exposure, the cause of loss, and the loss frequency and severity. financial consequences may be more difficult to determine, such as the value of business lost while the building damaged by fire is being restored.

Types of Loss Exposures

Property Loss exposures : A condition that presents the possibility that a person or an organizations will sustain a loss resulting from damage (including destruction, taking, or loss of use) to property in which that person or organization has a financial interest.

Property can be categorized as either tangible or intangible property.

1. Tangible Property : Property that a physical form.
2. Real Property (realty)  : Tangible property consisting of land, all structures permanently attached to the land, and whatever is growing on the land. 
3. Personal Property : All tangible or intangible property that is not real property.
4. Intangible Property : Property that has no physical form such as patents, copyrights, trademarks, trade secrets and customer goodwill.

Maximum financial consequence of a property loss is limited by the value of the property. However a property loss may also have an effect on the financial consequences of liability, personnel, or net income.

Liability Loss Exposures

Any condition or situation that presents the possibility of a claim alleging legal responsibility of person or business for injury or damage suffered by another party.

Insurers professional often use the rem "loss" to mean the event itself. In addition, they often refer to the loss in terms of the applicable property, the cause of loss, the consequences, or the applicable policy.

• When focusing on the type of property, they often refer to a "building loss" or a "personal property loss", regardless of the peril involved.
• When focusing on causes of loss, they often refer to a "fire loss", a "smoke loss" or a "theft loss"
• When focusing on consequences, they often refer to a "business income loss", an "extra expense loss", or an additional living expense loss" regardless of the type of property or causes of loss involved.
•  When focusing on the applicable policy, they often use the policy name or type, such as a "homeowners loss", an "auto loss", or a "business interruption loss".

In liability loss exposures, extra costs include, defense costs, other claim-related expenses, and potentially adverse publicity, all of which produce a financial loss.

Personnel Loss Exposure

A condition that resents the possibility of loss caused by a key person's death, disability, retirement, or resignation, that deprives an organization of the person's special skill or knowledge that the organization cannot readily replace.

Personal Loss Exposure (aka human loss exposure)

Any condition or situation that presents the possibility of a financial loss to an individual or a family by such causes as death, sickness, injury or unemployment. Example : Death of primary wage earner of a family.

Net income Loss Exposure

A condition that presents the possibility of loss caused by a reduction in net income or increase in expenses resulting from propoerty loss. An indirect loss is a loss that results from, but is not directly cause by, a particular cause of loss. Estimating indirect losses is often challenging because of the difficulty in projecting the effects that a direct loss will have on revenues or expenses.

In the insurance industry, the term "net income losses" is usually associated with property losses, and some insurance policies provide coverage for net income losses related to property losses. However, there are many other causes of net income losses. Some net income losses are associated with the liability or personnel loss exposures that have traditionally been the focus of risk management.

Besides these, other potential net income losses that may affect individuals or organizations include these:

• Loss of goodwill
• Failure to perform
• Missed opportunities

Reducing the Financial Consequences of Risk

• The overall financial consequences of risk for a given asses or activity is the sum of three costs:
• the expected cost of the value of lost because of actual events that cause a loss (expected loss frequency times expected loss severity).
• the cost of the resources devoted to risk management for that asses or activity (which includs the costs of loss control, loss financing, and risk reduction.
• the cost of residual uncertainty (which includes the effects of uncertainty on the proces of the firm's products and on the price of the firm's stock).

For a particular asset or activity , the cost of risk can be broken down this way:

• Cost of losses not reimbursed by insurance or other external sources
• cost of insurance premiums
• cost of external funds, such as interest payments on loans or the transaction costs of noninsurance indemnity.
• cost of measures to prevent or reduce the size of potential funds
• cost of implementing and administering risk management

Benefits to Individuals

• Purchasing auto liability insurance enables them to transfer to liability loss exposure to the insurer
• The second benefit of risk management for individuals is that if reduces the residual uncertainty associated with risk.

Benefits of organizations

• Organization usually choose to manage their risks, because they, too, benefit from preserving their financial resources.
• Preservation of financial resources adds value to the organization and makes it a safer and more attractive investment
• The protection that risk management affords an organization's financial resources can, in turn, provide confidence that capital is protected against future costs such as property loss, interruption of future income, liability judgement s, or loss of key personnel. This sense of confidence is attractive both to suppliers and customers.
• Risk management also can reduce the deterrence effect of risk; that is, it can improve an organization's capacity to engage in business activities by minimizing the adverse effects of risk.

Benefits to Society

• Increase to productivity within an economy and improve the overall standard of living. Improves the allocation of society's scarce resources.

Risk Management Program Goals

Senior management support is essential to an effective and efficient risk management program. To gain that support, a risk management program should promote the organization's overall goals.

Risk management makes those who own or run an organization more willing to undertake risky activities.

Pre-Loss Goals

Goals to be accomplished before a loss, involving social responsibility, externally imposed goals, reduction of anxiety, and economy. It should help ensure that the organization’s legal obligations are satisfied.

Operational goals include 

• Economy of operations : By comparing its costs oor (deparmental or organizational effeciency) of risk management with other similar organizations, an organization can measure its pre-loss goal of economy of operations.
• Tolerable uncertainty : To keep managers uncertainty at tolerable levels, how much money can company afford to loose. The goal of tolerable uncertainty is to allow managers to make and implement decisions without being unduly affected by uncertainty
• Legality
• The legality be based on :
• Standard of care that is owed to others
• Honor contracts entered into by the organization
• Federal, State, and Local Laws and Regulations
• Respond to liability exposures.
• Illegality is itself a loss exposure.
• Social Responsiblity, it is both a pre-loss and post-loss goal. maintain a good public image.

Post-Loss Goals

Risk Management program goals that should be in place in the event of significant loss. Post-loss goals on the operating and financial conditions that the organization's senior management would consider acceptable after a significant cant foreseeable loss. These are six possible post-loss goals:

• Survival : most basic goal is survival (resume operations) after a loss has occurred , while the mose ambitious goal in uninterrupted growth.
• Continuity of operations
• Such Steps include these:
• Identify activitites whose interruptions cannot be tolerated
• Identify the types of events that could interrupt such activities
• Determine the standby resources that must be immediately available to counter the effects of those losses
• Ensure the availability of the standby resources at even the most unlikely and difficult times.
• The last step, ensuring the availability of standby resources, is likely to add to an organizaiton's expenses, and, accordingly, achieving the continuity of operations goal trands to be more costly tht the more basic goal of survival.
• Profitability
• In a for profit, it is to generate a net income
• In a not-for-profit it is to operate within the budget
• Maintain established minimum amount of profit
• Such a program stresses insurance and non-insurance transfers, raising the total risk management and financing costs. 
• Earnings stability : After survival - earnings stability and profitability is secondary objective.
• Social Responsibility : Major goal for public entitites.
• Growth : Most ambitous after a loss has occured, but compared to survival it is difficult to achieve and sometimes conflicting.

The more ambitious a particular post-loss goal, the more difficult and costly it is to achieve.

Conflicting Goals : Spending money to reduce risk to a tolerable level, to meet legal obligations or to assure profitable growth can come at the expense of economizing operations (unprotected growth) or increase Risk Management costs to protect expanding resources (protected growth). 

• More ambitious post-loss goals are costlier.
• among pre-loss goals, all legalities must be met reagrdless of cost, but uncertainy and social responsibility must be comprimised with economy.
• Any other conflicts among goalsmust be reconciled by the RM professional.

The Risk Management Process

Step 1: Identifying Loss Exposures methods such as document reviews , compliance analysis and outside expertise.

Step 2: Analyzing Loss Exposures : considering frequency , severity, total dollar losses and timing of loses. Timing of losses is the interval betweenloss occurence and loss payment. Timing is important because model held in reserve to pay for a loss can be invested until the payment is made. 

Step 3: Examining the Feasibility of Risk Management Techniques : Risk Control Techniques and Risk Financing. These techniques are not used isolation.

Risk control techniques alter (minimize) the frequency and severity of losses, they also make losses more predicatble, and risk financing techniques pay for losses despite the controls.

Step 4: Selecting the Appropriate Risk Management Techniques or combination of risk management techniques 

• Risk Control Techniques
• Loss Prevention reducing the frequency of a particular loss
• Loss Reduction reducing the severity of a particular loss
• Separation invovlves a particular acticity or asset over several locations
• Duplication involves relying on backups, that is, spares or duplicates, used only if primary assets or activities suffer loss.
• Diversification involves providing a range of products and services used by a variety of customers.
• Risk Financial Techniques based on cost/benefit analysis
• Retention involves generalizing finds from within the organization to pay for losses
• Transfer invovlves generalizing funds frin outside the organization to pay for losses and includes insurance and noninsurance transfer
• The three forecasts a financial analysts of a risk management technique may be based on are these:
• A forecast of the dimensions of expected losses (frequently, severity, timing of payment, and total dollar losses).
• A forecast, for each feaible combination of risk management techniques, of the effect on the frequency, severity and timing of these expected losses.
• A forecast of the after-tax costs invovlved  in applying the various risk management techniques.

Step 5: Implementing the Selected Risk Management Techniques : The risk management techniques selected by for-profit organizations should be both effective in meeting the organizations' goals and economical.

Step 6: Monitoring Results and Revising the Risk Management Program : to adjust it to accommodate changes in loss exposures and the availability or cost-effectiveness of alternate risk management techniques.

1. Establishing standards of acceptable performance
2. Comparing actual results with these standards
3. Correcting substandard performance or revising standards that prove to be unrealistic
4. Evaluating standards that have been substantially exceeded
5. Reavaluate your prior decisions in light of actual results and new conditions and adjust as needed. 

Chapter 2 : Risk Assessment

1. Document Analysis : Standardized Documents, Company-Specific, reviewing multiple documents is necessary to avoind failing to identify important loss exposures.
2. Compliance Review
3. Inspections
4. Expertise within and beyond the organization

Risk Assessments Questionnaires and Checklists

Standardized checklists are published by America Management Association (AMA), the International Risk Management Institute (IRMI), the Risk and Insurance and Management Society (RIMS), and others.

Although some organizations, or trade associations have developed specialized checklists or questionnaires for their members most are created by insurers or insurance companies because

• to identify insurable hazard risks
• focus on listing the organization's assets, whereas others focus on identifying potential causes of loss that could affect the organization.

Linking loss exposures with the goals they support can be useful in analyzing the potential financial consequences of loss.

Checklists typically capture less information then questionnaires. They do not show how those loss exposures support or affect organization goals.

A Questionnaire captures more descriptive information than  a checklist. They can capture the amounts or values exposed to loss. It can also be designed to include questions that address key property, liability, net income, and at least some personnel loss exposures. Additionally, the logical sequencing of question helps in developing a more detailed examination of the loss exposures an organization faces.

Both checklist and questionnaires may be produced by insurers and are called insurance surveys. Most of the question helps in developing a more details examination for which commercial insurance is generally available.

Risk Management and Risk Assessment questionnaires have a broader focus and address both insurable and uninsurable loss exposures. However a disadvantage of risk assessment questionnaires is that they typically can be completed only with considerable expense, time, and effort and still may not identify all possible loss exposures.

Standardizing a survey or questionnaire has both advantage and disadvantages. Standardized questionnaires are relevant for most organization can can be answered by persons who have little risk management expertise. However, no standardized questionnaire can be expected to uncover all the loss exposures particularly characteristic of a given industry. They respondent might not do anything more then answer the questions and not reveal any key information.

The questions should ideally by used in conjunction with other identification and analysis methods. Experienced insurance and risk management professionals often follow up with additional questions that are not on the standardized document.

Financial Statement and Underlying Accounting Records

It can help in identify any future plans that could lead to new loss exposures. Helps to identify major causes of loss exposures.

Balance Sheet : The financial statement that reports the assets, liabilities, and owners' equity of an organization as of a specific date. Owners' equity, or net worth, is the amount by which assets exceed liabilities. Asset entries indicate property values that could be reduced by loss. Liability entries show what the organization owes and enable the risk management professional to explore two types of loss exposures:

1. Liabilities that could be increased or created by a loss
2. Contracts or Obligations (such as mortgage payments) that the organization must fulfill, even it it were to close temporarily as a result of a business interruption.
3. Property asset exposure can be seen in the asset section of the balance sheet.

Income Statement : The financial statement that reports an organization's profit or loss for a specific period by comparing the revenues generated with the expenses incurred to produce those revenues. Helps to identify net income loss exposures that reduce revenue or increase expenses.

Statement of Cash Flows : The financial statement that summarizes the cash effects of an organization's operating, investing and financing activities during a specific period. Using cash flow analysis can identify the amounts of cash either subjects to loss or available to meet continuing obligations. Expose financial risks, such as in the value of investments, interest rate, volatility, foreign exchange rate changes, or commodity price swings.

Major Disadvantage of financial statements for identifying loss exposures is that although they identify most of the major categories of loss, they do not identify or quantify individual personnel loss exposure. Another is the financial statement depict past activities but are of limited help in identifying projected values of future events.

Contracts

A contract is an agreement entered into by two or more parties that specifies the parties' responsibilities to one another. It is often necessary to consul legal experts when interpreting contracts.

Contract analysis can both identify the property and liability loss exposures generated or reduced by an organization's contract and ensure that the organization is not assuming liability that is disproportionate to its stake in the contract. Ongoing contract management is part of monitoring and maintaining a risk management program.

Two ways of contract liability 

1) Hold Harmless Agreement( or Indemnity Agreement ): A contractual agreement that obligates one of the parties to assume the legal liability of another party.

2) If the organization fails to fulfill a valid contract them it will be a liability loss exposure.

Insurance Policies

Insurance is a means of risk financing, reviewing insurance policies can also be helpful in risk assessment.

Identify potential loss exposures which is not covered in the insurance policies.

Risk Manager can compare his or her coverage against an industry checklist of insurance policies currently in effect.

Organization Policies and Records

To identify loss exposures such as work related hazards. using organizational policies and records, such as corporate by-laws, board minutes, employee manuals, procedure manuals, mission statements, and risk management policies.

Sometimes Internal Documents are also used to identify potential liability loss exposures.

One drawback is the sheer volume of records/document that the organization generate internally to identify loss exposures,therefore management professionals would need to examine a representative sample of documents. This may make the task manageable, but increases the likelihood that not all loss exposures will be identified.

Flowcharts and Organizational Charts

A flowchart is a diagram that depicts the sequence of activities performed by a particular organization or process. An organization can use flowcharts to show the nature and use of the resources involved in this operations as as well as the sequence of and relationships between those operations.

Organization indicates the hierarchy of an organizations personnel and can help to identify key personnel for whom the organization may have a personnel loss exposure. IT can also help in identifying bottlenecks that may exist.

Disadvantage : the personnel identified is not guaranteed that he/she is a key personnel and the relative importance of the individual to the continued operation of profitability of the organization.

Loss Histories

Based on past loss histories are important indicators of current or future loss exposures and may not identify loss exposures which may not have occurred in the past.

Compliance Review 

A compliance review determines an organizations' compliance with local, state, and federal statutes and regulations. The benefit of compliance reviews is the hey can help an organization minimize or avoid liability loss exposures because non-compliance can be a liability loss exposure.

Drawback is that they are expensive and time consuming. It is done by in-house legal and accounting resources otherwise it may have to sue outside expertise.

Personal Inspections

Loss exposures are best identified by personal inspections that would not appear on written descriptions of the organization's operations.

It should ideally be conducted by individuals whose background and skills, equip them to identify unexpected, but possible, loss exposures. The front-line personnel are best placed to identify non-obvious loss exposures.

Expertise within and beyond the Organization

Interviews should include a range of employees from every level of the organization. One area of specialization that often requires expertise from expert services is Hazard analysis which is a method that identifies conditions that increase frequency or severity of loss.

Data requirement for exposure analysis

Relevant Data : depends on the type of loss exposure studied

Complete Data : depends on the nature of loss exposure studied, having complete helps to isolate the nature of each loss

Consistent Data :  to identify past patterns of loss, so that future loss are not underestimated or overestimated.

1. the loss data must be collected on a consistent basis for all recorded losses
2. data must be expresses in constant dollars, to adjust for differences in price levels

Nominal Dollar - dollar values at the time of loss

Current Dollars - dollars value today

Real or Constant Dollars - dollar values in some base year. the value enables comparison of losses that have occurred in different time periods.

Organized Data : Organize losses by size is also the foundation for developing loss severity distributions or loss trends over time.

Nature of Probability : the probability of an event is the relative frequency with which the event can be expected to occur in the long run in a stable environment. determine the probability that a certain event occur can be an important part of exposure analysis in the risk management process.

Theoretical Probability : Probability that is based on theoretical principles rather than on actual experience. Insurance rarely use them because very few loss exposures can be modeled with theoretical, but it is a great starting point to model empirical probability or improve the model. can be represented with table, chart or graph.

Empirical Probability : A probability measure that is based on actual experience through historical data or from the observation of facts.

The empirical probability deduced solely from historical data may change as new data are discovered or as the environment that produces those event changes.

Empirical probabilities are only estimates whose accuracy depends on the size and representatives nature of the samples being studied. In contrast, theoretical probabilities are constant as long as the physical conditions that generate then remain unchanged.

constructed the same way as theoretical probability , 

1. first requirement is that it provides mutually exclusive and collectively exhaustive list of outcomes, loss categories (bins) must e designed so that all the losses can be included. One method is to divide the bins into equal number of sizes
2. define the set of probabilities of each possible outcome

Law of Large Numbers : 

A mathematical principle stating that as the number of similar but independent exposure units increase, the relative accuracy of predictions about future outcomes(predictions) also increases.

Limitation of law of large numbers

• the events have occurred in the past under substantially identical conditions and have resulted from unchanging, basic causal forces.
• the events can be expected to occur in the future under the same, unchanging conditions
• the events have been, and will continue to be, both independent of one another and sufficiently numerous.

Probability analysis : A technique for forecasting events, such as accidental and business losses, on the assumption that they are governed by an unchanging probability distribution which means

1. a substantial volume of data on past losses 
2. fairly stable operation so that (except) for price level changes) patterns of past losses presumable will continue in the future

In organizations with this type of unchanging environment, past losses can be viewed as a sample of all possible losses that organization will suffer.

Probability distribution can be constructed from empirical probabilities.

a presentation (table, chart, or graph, of probability estimates of a particular set of circumstances and of the probability of each possible outcome

Should be mutually exclusive and collectively exhaustive. Can be of discrete (loss frequency) and continuous type of distribution (loss severity).

discrete probability can be shows  as a table, example frequency distributions

continuous probability can be shows as a graph or by dividing the distribution into a countable number of bins. likelihood of those outcomes are plotted in a continuous graph and called as probability density functions. example : severity distribution.

the measures of central density represents the best guess as to what the outcome will be.

Central Tendency : the single outcome that is the most representative of all possible outcomes included within a probability distribution. can be used to compare the characteristics of different probability distributions.

Expected Value : the weighted average of all the possible outcomes of a theoretical, continuous and discrete probability distribution. for continuous it can be done but more difficult

Mean : for empirical distribution the sum of the values in a data set divided by the number of values. helps to give best guess of number of future events o dollar amount

Median : the value at the midpoint of a sequential data set with an odd number of values, or the mean of the two middle values of a sequential data set with an even number of values. a probability distribution median has a cumulative probability of 50 percent.

helps in selecting lower limits for retention levels and upper limits of insurance coverage

Mode : the most frequently occurring value in a distribution. Outcome directly below the peak of probability density function. Helps insurance companies to focus on outcomes that are most common.

In a symmetrical or standard bell-shaped distribution, the mean and the median has the same value.

Many loss distributions are skewed, asymmetrical distributions are common for severity distributions where most losses are small but there is a small probability of a large loss occurring. In such cases the median is a better guess than the mean.

Dispersion : used for analyzing probability distribution, to assess the credibility of the measures of central tendency in analyzing loss exposures.

Less dispersion means less uncertainty, will means less risk is involved in the loss exposure.

Measures of Dispersion 

1. Standard Deviation 
2. Coefficient of variation

A measure of dispersion between the values in a distribution and the expected value (or mean) of that distribution, calculated by taking the square root of the variance. 

Standard Deviation of theoretical probability distribution

1. Calculate the distribution's expected value or mean
2. Subtract the expected value from each distribution value to find the differences
3. Square each of the resulting differences
4. Multiple each square by the probability associated with the value 
5. Sum the resulting products
6. find the square root of the sum

Standard Deviation of Individual Outcomes - mostly used by insurance professional to measure dispersion of potential outcomes

1. calculate the mean of the outcomes ( the sum of the outcomes divided by the number of outcomes)
2. subtract the mean from each of the outcomes
3. square each of the resulting differences
4. sum these squares
5. divide the sum by the number of outcomes minus one( this value is called variance)
6. calculate the square root of the variance.

Coefficient of Variation

a measure of dispersion calculated by dividing a distribution's standard deviation by its mean. It is useful in comparing the variability of distributions that have different shapes, means, or standard deviation. the higher the variability, within a distribution, the more difficult it is to accurately forecast an individual outcome.

Normal Distribution

a probability that, when graphed, generates a bell-shaped curve.

1. 34.13 % of all outcomes are within one standard deviation above the mean
2. 68.26 % of all outcomes are within one standard deviation above and below the mean
3. 13.59 % of all outcomes are within one standard deviation and two standard deviation
4. 95.44 % of all outcomes are within two standard deviation above and below the mean
5. 99.74 % of all outcomes are within three standard deviation above and below the mean
6. 2.15 % of all outcomes are within two standard deviation and three standard deviation or 4.30%
7. 0.13 % are above three standard deviation or 0.26%

Using this analysis, risk management can select an appropriate probability that  level of coverage or risk

Analyzing Loss Frequency, Loss Severity, total dollar Losses and timing helps insurance and risk management professionals develop loss projections, and, therefore also helps them prioritize loss exposures so that risk management resources are prioritized for those loss exposures.

• Loss Frequency - The number of losses that occur during a specific period.
• Loss Severity - The dollar amount of loss for a specific period
• Total Dollar Losses - The total dollar amount of losses for all occurrences during a specific period. or maximum possible loss (MPL) is the total value exposed to loss at any one location or from any one event.
• Timing - The points at which losses occur and loss payments are made

If any of these dimensions of loss exposure analysis involve empirical distributions developed from past losses, the credibility of the data being used needs to be determined. Data credibility of the data being used needs to be determined which is the level of confidence that available data are accurate indicators of future losses

In liability cases, the maximum possible loss is limited to defendant's personal wealth. therefore, some practical assumptions must be made about the MPL in liability cases to properly asses that loss exposure. Instead of focusing on the defendant's total wealth, a common assumption is that the maximum amount that would be exposed to liability loss 95 percent (or 98 percent) of the time in similar cases is the MPL.

To study and prioritize loss frequency and loss severity joint the Prouty Approach,identifies the below categories

Loss Frequency - 

Almost nil - Extremely unlikely to happen, virtually no possibility

Slight - Could happen but has not happened

Moderate - Happens occasionally

Definite - Happens regularly

Loss Severity -

Slight -  Organization can readily retain each loss exposure

Significant - Organization cannot retain the  loss exposure, some part of which must be financed

Severe - Organization must finance virtually all of the loss exposure or endanger its survival

Loss frequency and loss severity tend to be inversely related. The more severe a loss tends to be, the less frequently it tends to occur. conversely, the more frequently a loss occurs to a given exposure, the less severe the loss tends to be.

Another to study loss frequency and severity is to combine both into a total claims distribution.

Expected Total Dollar Losses can be projected by multiplying expected loss frequency by expected loss severity, while worst-case scenarios can be calculated by assuming high frequency and the worst possible severity.

Chapter 3 - Risk Control  

Risk control : A conscious act or decision not to act that reduces the frequency and severity of losses or makes losses more predictable.

Risk control techniques can be classified using these six broad categories : 

1. Avoidance
2. Loss Prevention
3. Loss Reduction
4. Separation
5. Duplication
6. Diversification

Generally a loss prevention measure is implemented before a loss occurs in order to break the sequence of events that leads to the loss.

Heinrich's domino theory related to work injuries  : included a sequence of five dominoes of which if any one can be removed then the loss can be prevented.

These are 

1. social environment and ancestry
2. the fault of person
3. personal or mechanical hazards
4. the accident
5. the injury

Loss Reduction : A risk control technique reduces the severity of a particular loss. Some loss reduction measures can prevent losses are well as reduce them.

The two broad categories of loss reduction measures are pre-loss measures, applied before the loss occurs, and post-loss measures, applied after the loss occurs. The aim of pre-loss measures, is to reduce the amount or extent of property damaged and the number of people injured or the extent of injury from a single event.

Post Loss measures typically focus on emergency procedures, salvage operations, rehabilitation activities, public relaxations, or legal defenses to halt the spread or to counter the effects of loss.

Disaster Recovery Plan : is a specialized aspect of loss reduction, also called catastrophe recovery plan or contingency plan, is a plan for backup procedures, emergency response, and post-disaster recovery to ensure that critical resources are available to facilitate the continuity of operations in an emergency situation without which the organization could not functions. Disaster Recovery plans typically focus on loss property loss exposures and natural hazards, not on the broader array of risks and associated loss exposures that may also threaten an organization's survival.

Separation : A risk control technique that isolates loss exposures from one another to minimize the adverse effect of a single loss.

It is rarely undertaken by its own sake, but is usually a byproduct of another management decision. The intent of separation is to reduce the severity of an individual loss at a single location. However, by creating multiple location, separation more likely increases loss frequency.

Duplication : A risk control technique that uses backups, spares, or copies of critical property, information, or capabilities and keeps them in reserve.

Duplication differs from separation in that duplicates are not a part of an organization's daily working resources Duplication is only appropriate if an entire asset or activity is so important that the consequences of its loss justifies the expense and time of maintaining the duplicate. Like separation, duplicate can reduce an organization's dependence on a single asset, activity or person, making individual losses smaller by reducing the severity of a loss that may occur. Duplication is not as likely as separation to increase loss frequency because the duplicates unit is kept in reserve and is not as exposed to loss as is the primary unit.

Duplication is likely to reduce the average expected annual loss from a given loss exposure because it reduces loss severity without increasing loss frequency significantly. Similar to separation, duplication can also make losses more predictable by reducing the dispersion of potential losses.

One option is for an organization to contractually arrange for the acquisition of equipment or facilities in the event that a loss occurs.

Diversification : A risk control technique that spreads loss exposures over numerous projects, products, markets, or regions.

Organization engage in diversification of loss exposures when they provide a variety of products and services that are used by a range of customers

As with separation, diversification has the potential to increase loss frequency, but by spreading risk , diversification reduces loss severity and can make losses more predictable.

Risk Control Goals

• Implement effective and efficient risk control measures - pre-loss and post-loss goals. Some risk control measures will be more effective than others, effectiveness is based on both quantitative and qualitative standards.
• Comply with legal requirements : such as federal statute, and support pre-loss goal of legal liability, part of the cost of risk
• Promote life safety 
• Ensures business continuity

thereby support risk management program and helping organization achieve its goals.

Several measures of comparison of effectiveness, one of them is cash-flow analysis. the major advantage of using cash flow analysis for selecting risk control measures is that it provides the same basis of comparison for all value-maximizing decision and thereby helps the organization achieve its value-maximization goal. It is also very useful for not-for-profit organizations that want to increase their efficiency by reducing unnecessary expenditures on risk control.

The disadvantage of cash flow analysis include the weakness of the assumptions that often must be made to conduct the analysis and the difficulty of accurately estimating future cash flows. Moreover cash flow analysis works on the assumption that the organization's only goal is to maximize its economic value and does not consider any of the non-financial goals or selection criteria.

Life Safety : The portion of fire safety that focuses on the minimum building design, construction, operation, and maintenance requirements necessary to assure occupants of a safe exit from the burning portion of the building. these standards are codified in the Life Safety Code published by the National Fire Protection Association (NFPA) and cover the risk control technique of avoidance, loss prevention, and loss control.

Business Continuity : Business Continuity is designed to meet both the primary risk management program post-loss goal of survival and post-loss goal of continuity of operations. Loss exposures and their associated losses vary widely by industry, location, and organization. Because each organization is unique in its potential losses, each must also be unique in its application of risk control measures to promote business continuity

Applications of Risk Control Technique

Applicable to each of the below exposure

• Property
• Liability
• Personnel
• Net Income

Property loss exposures are generally divided into two categories - tangible and non-tangible. The risk control techniques that are most applicable to property loss exposures vary based on the type of property as well as the cause of loss threatening the property.

insurance producers and underwriters commonly examine commercial property loss exposures based on construction, occupancy, protection, and external exposure (known by their acronym - COPE - Construction, Occupancy, Protection, External Exposure).

Liability losses basis can be legal actions that can be brought by torts, contracts, or statutes. Three risk control techniques can be used to control liability losses

1. Avoid the activity that creates the liability loss exposures
2. decrease the likelihood of the losses occurring (loss prevention)
3. if a loss does occur, minimize its effect on the organization (loss reduction)

Loss Prevention and Loss Reduction are most commonly used.

The most common loss prevention measure is to control hazards (conditions that increase the loss frequency or severity).

After a liability loss has occurred, individuals and organizations can implement loss reduction measures to reduce the severity of the liability loss. Such measures can include these:

1. Consulting with an attorney for guidance through the legal steps necessary to resolve liability claims.
2. Property responding to the liability claim and to the claimant in order to avoid feelings of ill will that may increase the claimant's demands.
3. Participation in alternative dispute resolution. Litigation is a long and costly process. Some forms of alternative dispute resolution, such as mediation or arbitration, often help to resolve liability claims more quickly and more economically than litigation.

Personnel Loss Exposures

Loss prevention measures used to control work-related injury and illnesses typically involve education, training, and safety measures. An organization may also attempt to prevent personnel causes of loss that occur outside workplace by controlling key employees' activities through employment contracts. alternatively organizations may see a form of separation, such as restricting the number of key employees who can travel on the same aircraft.

Although all organization must comply with federally mandated safety measures issued by OSHA (the Occupational Safety and Health Administration), additional training and safety precautions are often cost-effective.

Net Income Loss Exposures : All  measures to above exposures indirectly control net income loss. In addition to reducing the immediate effect property,liability or personnel losses, risk control effects must also control long-term effects such as a loss of market share that can result from the net income loss. Two risk control measures aimed at reducing the severity of losses is duplication and separation. Diversification is a also a viable risk control technique.

Business Continuity Management

Initial focus was on IT but now business continuity management plans have been expanded to 

1. Property Losses
2. IT Problems
3. Human Feelings
4. Loss of Utility services or infrastructure
5. Reputation losses
6. Human asset losses (personnel losses)

A systematic approach to developing and implementing a business continuity plan involves these six steps :

1. Identify the organization's critical functions
2. Identify the risks( threats) to the organization's critical functions
3. Evaluate the effect of the risks on those critical functions
4. Develop a business continuity strategy
5. Develop a business continuity plan
6. Monitor and revise the business continuity process

These steps are similar to risk management process and are designed to assess and control risks

The duration of interruption necessary to produce a substantial effect depends on the function. It can help to minimize the scope of threats that can cause its demise.

A Business Continuity plan details the activities the organizations will take in response to an incident that interrupts its operations. The plan should be designed with the understanding that it is going to be used during a crisis; that is, it should be clear and able to be quickly reach and understood. It also provides a framework for organizations to develop a systematic response to a variety of risks that could be potentially threaten the future viability of the organization..

business continuity management and risk control are aimed at enabling an individual or organization to not only deal with hazards and loss exposures, but deal with them in the most efficient and cost-effective way in order to reduce the exposures to, and cost of, risk.

Chapter 4 : Risk Financing

Pay for Losses

Risk Financing measure should be effective (pay for losses that do occur) as well as efficient (pay for losses in the most economical way).

Recruiting at a startup is very different from hiring at a big company

The first thing you notice at a big company is the amount of specialization. At a startup, everyone does a little of everything, so you need strong generalists. More importantly, it’s hard to predict the future, so you need people who can adapt. You might think you’re hiring somebody to work on something specific, but that something might change in a few months. It doesn’t work that way at big companies. Usually when you’re hiring you have a very specific role in mind, and the likelihood that that responsibility will change is low.

Hire all the smart people

take a wickedly smart, inexperienced PM over one of average intellect and years of experience any day

Leadership that’s earned

Product managers are usually leaders in their organizations. But they typically don’t have direct line authority over others. That means they earn their authority and lead by influence

“Spidey-sense” product instincts and creativity

Strong technical background

Having a solid engineering background gives a PM two critical tools - the ability to relate to engineers and a grasp of the technical details driving the product.

Ability to channel multiple points-of-view

Being a product manager requires wearing multiple hats. That means you need to be capable of doing other people’s jobs, but smart enough to know not to. Great PMs know how to channel different points-of-view. They play devil’s advocate a lot

Give me someone who’s shipped something

This last characteristic may be the easiest to evaluate. Unless the position is very junior, I’ll usually hire product managers who’ve actually shipped a product. I mean from start to finish, concept to launch. Nothing is a better indication of someone’s ability to ship great products than having done it before. Past performance is an indication of future success. Even better, it gives something tangible to evaluate in a sea of intangibles

Good Anecdote from HBR review : Machine learning excels at predicting things. It can inform decisions that hinge on a prediction, and where the thing to be predicted is clear and measurable.

Power of Analytics - Predictive Analytics is about tomorrow

1. Study One Factor — Using basic spreadsheet software, study historic trends in your business to forecast expected revenue tomorrow, next week, or next year, which is useful for setting budgets and goals. Data scientists call this kind of analysis “univariate time series” because you look at only one variable over time, ignoring how other factors might come into play. For example, you might look at the timing of offers you have made and how well they have done.

Hype around machine learning 

Explore more problems: Data scientists need the ability to rapidly define and explore multiple prediction problems, quickly and easily. Instead of exploring one business problem with an incredibly sophisticated machine learning model, companies should be exploring dozens, building a simple predictive model for each one and assessing their value proposition.

Learn from a sample of data-not all the data: Instead of focusing on how to apply distributed computing to allow any individual processing module to handle big data, invest in techniques that will enable the derivations of similar conclusions from a data subsample. By circumventing the use of massive computing resources, they will enable the exploration of more hypotheses.

Focus on automation: To achieve both reduced time to first model and increased rate of exploration, companies must automate processes that are normally done manually. Over and over across different data problems, we found ourselves applying similar data processing techniques, whether it was to transform the data into useful aggregates, or to prepare data for predictive modeling—it’s time to streamline these, and to develop algorithms and build software systems that do them automatically.

For example, marketers often compare customer lifetime value with the cost of acquiring a customer. The problem is that customer lifetime value relies on a prediction of the net profit from a customer (so it’s largely unobserved and uncertain), while the business has much more control and certainty around the cost of acquiring a customer (though it’s not completely known). Treating the two values as if they’re observed and known is risky, as it can lead to major financial losses.

1 – Risk management

No matter the industry, Insufficient risk analysis is often a leading cause of failure. This is especially true in the financial industry where adoption of Risk Management Software based on text mining technology can dramatically increase the ability to mitigate risk, enabling complete management of thousands of sources and petabytes of text documents, and providing the ability to link together information and be able to access the right information at the right time.

2 – Knowledge management

Not being able to find important information quickly is always a challenge when managing large volumes of text documents—just ask anyone in the healthcare industry. Here, organizations are challenged with a tremendous amount of information—decades of research in genomics and molecular techniques, for example, as well as volumes of clinical patient data—that could potentially be useful for their largest profit center: new product development.  Here, knowledge management software based on text mining offer a clear and reliable solution for the “info-glut” problem.

3 – Cybercrime prevention

The anonymous nature of the internet and the many communication features operated through it contribute to the increased risk of  internet-based crimes. Today, text mining intelligence and anti-crime applications are making internet crime prevention easier for any enterprise and law enforcement or intelligence agencies.

4 – Customer care service

Text mining, as well as natural language processing are frequent applications for customer care. Today, text analytics software is frequently adopted to improve customer experience using different sources of valuable information such as surveys, trouble tickets, and customer call notes to improve the quality, effectiveness and speed in resolving problems. Text analysis is used to provide a rapid, automated response to the customer, dramatically reducing their reliance on call center operators to solve problems. 

5 – Fraud detection through claims investigation

Text analytics is a tremendously effective technology in any domain where the majority of information is collected as text. Insurance companies are taking advantage of text mining technologies by combining the results of text analysis with structured data to prevent frauds and swiftly process claims.

6 – Contextual Advertising

Digital advertising is a moderately new and growing field of application for text analytics. Here,  companies such as Admantx have made text mining the core engine for contextual retargeting  with great success. Compared to the traditional cookie-based approach, contextual advertising provides better accuracy, completely preserves the user’s privacy.

7 – Business intelligence

This process is used by large companies to uphold and support decision making. Here, text mining really makes the difference, enabling the analyst to quickly jump at the answer even when analyzing petabytes of internal and open source data. Applications such as the Cogito Intelligence Platform (link to CIP) are able to monitor thousands of sources and analyze large data volumes to extract from them only the relevant content.

8 – Content enrichment

While it’s true that working with text content still requires a bit of human effort, text analytics techniques make a significant difference when it comes to being able to more effectively manage large volumes of information. Text mining techniques enrich content, providing a scalable layer to tag, organize and summarize the available content  that makes it suitable for a variety of purposes.

9 – Spam filtering

E-mail is an effective, fast and reasonably cheap way to communicate, but it comes with a dark side: spam. Today, spam is a major issue for  internet service providers, increasing their costs for service management and hardware\software updating; for users, spam is an entry point for viruses and impacts productivity. Text mining techniques can be implemented to improve the effectiveness of statistical-based filtering methods. 

10 – Social media data analysis

Today, social media is one of the most prolific sources of unstructured data; organizations have taken notice. Social media is increasingly being recognized as a valuable source of market and customer intelligence, and companies are using it to analyze or predict customer needs and understand the perception of their brand. In both needs Text analytics can address both by analyzing large volumes of unstructured data, extracting opinions, emotions and sentiment and their relations with brands and products.

Chapter 1 - Exploring Big Data Analytics

Technology has transformed our lives in how we communicate, learn and do business. The convergence of big data and technology has started a transformation of the property-casualty insurance business.

1. Data from internal source : Data that is owned by an organization, can include traditional and non-traditional data. Example: Text mining and Data Mining
2. Data from external source : Data that belongs to an entity other than the organization that wishes to acquire and use it.  Example: social media, statistical plan data, other aggregated insurance industry data, competitor's rate filings, telematics (driving patterns) and economic data and geodemographic data and IOT (Internet of Things)

Economic Data : Data regarding interest rates, asset prices, exchange rates, the Consumer Price Index, and other information about the global, the national, or a regional economy

Geodemographic data : Data regarding classification of a population.

Data science is especially useful to gather, categorize, and analyze unstructured data. new techniques to analyze, explore data. Even if the results may be useful or not

• Text Mining : Obtaining information through language recognition. Example : can be used to analyse claims adjusters' notes to identify fraud indicators.
• Data Mining : The analysis of large amounts of data to find new amounts of data to find new relationships and patterns that will assist in developing business solutions.This technique can be used to identify previously unknown factors that are common to an insurer's most profitable auto insurance customers. Also find innovative ways to market to those customers.
• Telematics : Evolving source of big data. The use of technological devices in vehicles with wireless with wireless communication and GPS tracking that transmit data to businesses and GPS tracking that transmit data to businesses or government agencies; some return information for the driver. This technique can be used to find driving patterns and change premiums dynamically. The information obtained from telematics can also be applied to an insurer's online application and selection process. Telematics is also an example of structured external data.

• First, analysis performed by computers is not always accurate. 
  
• Second, the automated analysis may be correct but irrelevant to a given business problem. 
  
• And third, just as the technology is rapidly evolving, so are the physical, political, economic, and business environments. 

• Data characteristics : 
• volume : size of the data
• variety : structured and unstructured data
• velocity : growing rate of change in the types of data
• veracity : completeness and accuracy of data
• value : goal of data science to derive value from the results of data analysis to help insurers make better decisions.
• Internal (example: risk factors, losses, premium, rating factors, rates and customer information) and External Data
  
• Structured data : data growing into databases with defined fields, including links between databases.
• Unstructured data : data that is not organized into predetermined formats, such as databases and often consists of text, images, or other non-traditional media, news reports

!!There is a sometimes a blurring of the boundary between internal and external data.!!

• The accident years included in the report
• The range of dates of paid loss transactions
• The evaluation data of the case-based loss reserve transactions
• the definitions of incurred losses (such as paid plus latest loss reserve, inclusive or exclusive of allocated expenses,net of deductible or reinsurance, and so forth).

Metadata can provide criteria for edits or restrictions of data inputs. A best practice is to reject, or at least flag, inputs that fail to meet the criteria. 

Metadata can be enhanced by documentation.!!

Examples of inaccurate data

1. Missing data and null values
2. Data errors in entering
3. Default values rather than actual values
4. Duplicate transactions

1. Quantitative or numerical
2. Categorical or alphabetic

Descriptive Statistics : quantitative summaries of the characteristics of a dataset, such as the total or average. can also be used to identify missing or inaccurate quantitative data and also outliers. The descriptive approach is applied when an insurer or risk manager has a specific problem.

Categorical Data : a multidimensional partitioning of data into two or more categories also called chunks or data cube. These chunks are then organized into tables that will allow the data to be analyzed for unusual values. A data cube is a multidimensional partitioning of data into to or more categories. It is sued to determine the percentage of each combination of injury type and training in the accident data for the past year. Data cubes can be used to identify inaccuracies and omissions in categorical data.

Data Security : Best practice to safeguard important data on which important decisions are based is to restrict the data so that no one can change the data. Authorized users' lack of access should be based on their responsibilities. and use cyber security techniques to reduce the likelihood that malware will corrupt data in their systems

Malware : Malicious software, such as a virus, that is transmitted from one computer to another to exploit system vulnerabilities in the targeted computer.

Data mining is closely related to the fields of statistics, machine learning and database management.

Statistics : A field of science that derives knowledge from data; it provides a root understanding of useful approaches to data analysis.

Database : A collection of information stored in discrete units for ease of retrieval manipulation, combination or other computer processing.

Algorithm : An operational sequence used to solve mathematical problems and to create computer programs.

Basic techniques of data mining :

• Classification : assigning members of a dataset into categories based on known characteristics.
• Regression Analysis : a statistical technique that predicts a numerical value given characteristics of each member of a dataset.
• Association Rule learning : examining how data to discover new and interesting relationships.for these relationships, algorithms are used to develop new rules to apply to new data.
• Cluster Analysis :  using statistical methods, a computer program explores to find groups with common and previously unknown characteristics. the results of the cluster analysis may or may not provide useful information.

Association rule learning and cluster analysis (exploratory techniques)are used to explore data to make discoveries. Insurers, like other business, might apply cluster analysis to discover customer needs that could lead to new products. Unlike classification and regression analysis, there are no known  characteristics of the data beforehand. The purpose of association rule learning and cluster analysis to to discover relationships and patterns in the data and then determine if that information is useful for making business decisions.

A predictive approach to data analytics involves providing a method to be used repeatedly to provide information.

Cross Industry Standard Process for Data mining (CRISP-DM) : An accepted standard for the steps in any data mining process used to provide business solutions. developed in 1999, a consortium of individuals who worked in the emerging field of data mining at DaimlerChrysler, NCR, and SPSS received funding from the European Union to develop a data mining process standard.

Steps in CRISP-DM

1. to understand what a business want to achieve by applying data mining to or more sets of data
2. types of data hat are being used - internal or external - structured or unstructured
3. cleaning or pre-processing of data
4. data mining techniques are applied to develop a model
5. results are evaluated to determine whether they are reasonable and meet the business objective
6. pre deployment or post deployment of the model, continuous refinement of the model is done to produce increasingly better and more accurate data. the circle the surrounds the diagram, indicate the data mining is a continuous process that involves continuously evaluating and refining the model.

A central feature of CRISP-DM is the circle that represents it is an ever evolving process

Data Science :

Is a new field at the frontier of data analytics. It is often experimental, and methods evolve rapidly. It uses the scientific method, which consists of these steps:

• a question or problem is raised
• Research is conducted regarding the subject
• a hypothesis is developed, based on the research, regarding the answer to the question or the cause of/solution to the problem.
• Experiments are performed to test the hypothesis
• Data from the experiments is analysed
• A conclusion is reached

The quest of data science, as with all the sciences, is to increase knowledge of the world and provide more advanced solutions to complex questions and problems.

!!Data science is especially useful for unstructured data.!!

Four fundamental concepts of data science:

1. Systematic process can be used to discover useful knowledge from data, this framework forms the foundation for data-analytical processes.
2. Information technology can be applied to big data to reveal the characteristics of groups of people or events of interest.
3. Analyzing the data too closely can result in interesting findings that are not generally applicable
4. The selection of data mining approaches and evaluation of the results must be thoughtfully considered in the context on which the results will be applied

Data Scientist : Traditionally data is usually numerical or categorical. Data Scientists must be able to analyze traditional data as well as new types, including texts, geo-location based data, sensor data, images, and social network data. They must also have the skills to manage increasingly large amounts of data.

Actuary : A person who uses mathematical methods to analyze insurance data for various purposes, such as to develop insurance rates or set claim reserve. focus primarily or pricing, rate-making or claim reserving.

Actuaries are the insurance professionals who have traditionally analyzed data and made predictions based on their analyses.

however there is no clear distinction between actuary and data scientist

Information related to the context for data science is often referred to as domain knowledge.

Risk management and insurance professionals can be valuable members of data science teams.

Insurance company uses data driven decision using traditional methods, however data analytics can improve the types of data, methods of analysis and results by applying data-driven decision modeling.Discovering new relationships in data is a way insurers and risk managers can use data science to improve their results through data-driven decision making.

Data-driven decision making : An organizational process to gather and analyze relevant and verifiable data and then evaluate the results to guide business strategies. there are two approaches predictive approach and descriptive approach

Improvements via data-decision making

1. Automating decision making for improved accuracy and efficiency
2. Organizing large volumes of new data
3. Discovering new relationships in data
4. Exploring new sources of data

Descriptive Approach : is applied when an insurer or risk manager has a specific problem. Data Science is intended to be used to provide data that will help solve the problem. Insurers or Risk Managers do not continue to use data-driven decision making beyond the specific problem. This is a one-time problem for data-driven decision making, and the analysis will not be repeated.

Predictive Approach : A predictive approach to data analytics involves providing a method that can be used repeatedly to provide information for data-driven decision making by humans, computers or both.

Approach for Analytics

• Analyze the data
• Improve data quality to make it accurate, appropriate, reasonable and comprehensive.
• Select the analytical technique
• Make a data-driven decision

In a data - driven decision making the below steps are followed:

1. clearly define the problem
2. select appropriate data
3. prepare the data by removing missing or inaccurate data
4. select relevant variables
5. develop a model
6.  to discover patterns and correlations in the data

Chapter 2 - Predictive Modeling Concepts

1. Supervised Learning : A type of model creation, derived from the field of machine learning, in which the target variable is defined, challenge is that data should be there about the target
2. Unsupervised Learning : A type of model creation, derived from the field of machine learning, that does not have a defined target variable. Can be used to pre-process data into groups before supervised learning is used. Can be used to provide the information needed to define an appropriate target for supervised learning.

• A disadvantage is that unsupervised learning can sometimes provide meaningless correlations
• conducting unsupervised learning first may provide the information needed to define an appropriate target for supervised learning

The below techniques are also used 

1. Predictive Model : A model used to predict an unknown outcome by means of a defined target variable. It can predict values in the future past and present.
2. Descriptive Model : a model used to study and find relationships within data. It can help in gaining insight.

Other terms used in modeling techniques:

• Attribute : A variable that describes a characteristic of an instance within a model
• Instance (example) : the representation of a data point described by a set of attributes within a model's dataset.
• Target Variable : The predefined attribute whose value is being predicted in a data analytical/predictive model.
• Class Label : The value of the target variable in a model. For example is the variable binary

After an insurance selects the business objectives of its business model and the data that will be analyzed, an algorithm is selected.

Machine learning algorithms can take a number of forms, such as mathematical equations, classification trees and clustering techniques. Experienced actuaries and data scientists have the experience needed to select which algorithm is appropriate for a particular problem.

Information Gain : A measure of the predictive power of one or more attributes. can be thought of in terms of how much if affects the entropy of a given dataset, which is also a indication of how it provides about the target variable.

Entropy : A measure of disorder in a dataset, it is essentially  a measure of how unpredictable some is.

When a dataset is segmented based on informative attributes (those with high information gain). the entropy decreases.

Lift : In model performance evaluation(or value of a model), the percentage of positive predictions made by the model divided by the percentage of positive predictions that would be made in the absence of the model. Helps in determining the value the model brings to the business.

1. compile a list of general attributes which indicate a desirable risk
2. using a majority combining function, the majority of nearest neighbors's values predict the class label of the target variable
3. to calculate the weighted average

Majority Combining function : the majority of nearest neighbors' values predict the class label of the target variable, it gives equal weight to all of the nearest neighbors. It does not consider their distance.

A majority combining function gives equal weight to all of the nearest neighbors, while a weighted average weights the nearest neighbors' contributions by their distance.

• A model's performance on the holdout data will not sufficiently reassure its developers it will perform well in production
• A very limited amount of training data may be available, and the model's developers think it unwise to not use some of the other data for training because of the need for holdout data. 

Chapter 3 - Big Data Analysis Techniques

Most insurance and risk management professional will not directly apply data analysis techniques. Knowledge and fundamental understanding of how big data interacts with technology and how it can provide new information that will offer a competitive advantage to insurers and to insurance professionals who want to advance their careers.

Traditional data analysis techniques continues to be important in insurance and risk management functions for below reasons :

1. These techniques remain important in providing information and solving business problems.
2. They form the foundation for the newer techniques. It is also important for professionals to have some understanding of the newer techniques to be able to work with data analysts in teams to better understand the future of insurance.
3. Also, these techniques can often be adapted or used in various combinations to analyse large volumes of data or data that comes from nontraditional sources, such as social media.
   

Unsupervised Learning : Is a model that explores data to find relationships that can be further analyzed. Example : Text mining of Social Media for new insurance products.

There is often an interesting relationship; for example after information is obtained through unsupervised learning, a predictive model could be developed with that information for supervised learning.

Exploratory Data Analysis : Is a valuable approach to a business problem that can be used before developing and testing a model. The analyst can also obtain information about missing or inaccurate data. The techniques involve charts, and graphs that show data patterns and correlations among data.

Examples of charts and graphs

1. Scatter Plot : which is a two-dimensional plot of point values and show the relationship between two attributes.
2. Bubble Plot : Is a type of scatter plot in which the size of the bubble represents a third attribute for which there is data.
3. Correlation Matrix : similar to a 2-dimensional matrix with cells, the stronger the correlation the darker the the cell color shade in the matrix

The above graphs and plots are used to examine relationships before developing a model

Data Analysis Technique : After exploratory data analysis is complete and a decision is made to develop a predictive model, the analyst selects the most appropriate technique for a model, that will fit the business context and the type and source of the data. Like data and technology in general, the techniques continually evolve.

Example of some techniques

1. Segmentation : An analytical technique in which data is divided into categories, can be used for both supervised and unsupervised learning.
2. Association Rule Learning : Examining data to discover new and interesting relationships among attributes that can be stated as business rules, involves unsupervised learning in which a computer uses various algorithms to find potentially useful relationships among large amounts of data and to develop a rule that can be applied to new data. Association rule learning is used to discover relationships between variables in order to develop a set of rules to make recommendations.

Use Case for association rule learning : Insurers may use association rule learning to recommend an umbrella policy with Auto and Homeowner's policy.

Traditional Data Analysis techniques are still used to solve business problems, such as determining rates for insurance products and reserves for unpaid or future claims. These techniques are usually applies to structured data, such as loss cost by claimant and type of injury. sometimes external structured data is also used, such as gross domestic product by month.

Use Case : One of the following types of outcomes through traditional data analysis

1. A non-numerical category into which data will belongs such as buckets or bins
2. A numerical answer based on data, a model which give a value such as linear regression
3. A probability score based on historical data, example would be catastrophe loss for commercial insurance , example model as classification tree or logistic regression
4. A prediction of future results based on current and past data, for example an insurer or a risk manager wants to predict the cost of employees' back injuries, we can use decision tree and event tree analysis

Traditional data analysis techniques include classification trees, various types of statistical regression models, and cluster analysis. Moreover these types of techniques have been applied using machine learning to improve the accuracy of results.

Classification Trees : A supervised learning technique that uses a structure similar to a tree segment data according to known attributes to determine the value of categorical target variable.

• Node  : A representation of a data attribute.

• Arrow : A pathway in a classification tree.

• Leaf node : A terminal node in a classification tree that is used to classify an instance based on its attributes. the leaf node determines the value of the target or output variable or probability or classification. The part of a classification tree that indicates the classification of the target variable is the leaf node.

It is important to understand that these classifications are not necessarily what the actual outcomes will be

Regression Models : To determine numerical value for a target variable

Linear Regression : A statistical method (or algorithm) to predict the numerical value of a target variable (which can also be a ratio of two attributes) based on the values of explanatory variables. Using averaging method to predict average output value of target variable. The working of linear regression algorithm id developed using a method that minimizes the errors represented by the difference between the actual target variable values in the data and those forecast by the model.

Generalized Linear Model : (should not be confused with general linear model, which is a broad group of different types of linear models). A statistical technique that increases the flexibility of a linear model by linking it with a nonlinear function. It is used for more complex data classification and is widely used in Property-Casualty insurance business. A generalized linear model contains three components :

• the first is a random component, which refers to the probability distribution of the response variable. 
• The second is a systematic component, which describes the linear combination of explanatory variable (attributes). 
• The third component is a link function, which relates the results of the random and systematic components. 
• Link function : A mathematical function that describes how the random values of a target variable depend on the mean value generated by a linear combination of the explanatory variables (attributes).

Cluster Analysis : is unsupervised learning. A model that determine previously unknown groupings of data

Data analytics has to be handle not only large volumes of data but also rapidly increasing velocity of data. Data generated from new technologies are often unstructured and therefore require new techniques for analysis.

Association Rule Learning : is used to discover relationships between variables in order to develop a set of rules to make recommendations.

Data generated from new sources technologies are often unstructured and therefore require new techniques for analysis.

Text Mining : Modeling approach to search for words that are neither rare nor too common. Another modelling approach is to look for adjacent words. Example : the words "serious" combines with "injury: would likely be indicators of claim severity.

Social Network Analysis : The study of the connections and relationships among people in a network. Analysis helps in finding preferences or associations in a social network (Email chain is also a social network) . Helps is determining claims fraud rings or probability of fraud.

Neural Network : Technique can be used for both supervised and unsupervised learning. A data analysis technique composed of three layers, including an input layer, a hidden layer with non-linear functions, and an output layer, that is used for complex problems. The neural network is able to model complex relationships between the input and output variables.

The combination of analytics and machine learning has enhanced insurers' predictive modeling capabilities.

Classification Tree model : A classification tree is then built, through an induction process, where the computer recursively analyzes different splits on the value of the attributes in various sequences based on training data. the tree is then applied to holdout data to see whether it generalizes so as to make accurate predictions using data that was not included in its development. There is a risk that the tree might overfit the training data. In such cases, it will not produce useful results and will need to be pruned.

When developing a classification tree model, various attributes are independently analysed for the information gain they provide in classifying

!! Note : the sequence of attributes does not follow the relative information gain ranking of the attributes, which is often the case when the attributes are analyzed recursively to construct a tree !!

Recursively : Successfully applying a model

Root Node : The first node in a classification tree that provides the greatest information gain for determining the value of the target variable.

Each classification rule is represented by a leaf (attribute) node of the classification tree.

Probability Estimation Tree : For more information users of predictive models want to know the predicted class for a new instance but also the likelihood, or probability, that it is correctly classified. This information enables them to do a cost-benefit analysis. the data to construct a classification tree can be used to estimate a probability. This is known as tree-based class probability estimation. Tree-based probabilities are calculated by dividing the number of times the model correctly predicted the value of the target variables by the number of total predictions for each class at each leaf node.

Linear functions to make business decisions

Linear function : A mathematical model in which the value of one variable changes in direct proportion to the change in the value of another variable.

Linear Discriminant : A data analysis technique that uses a line to separate data into two classes. Used to separate and classify the data. Miss-classifications can be evaluated by accuracy and precision of the model.

Regression Analysis : A statistical technique that is used to estimate relationships between variables.

Logistic Regression : A type of generalized linear model in which the predicted values are probabilities.

Instance space : An area of two or more dimensions that illustrates the distribution of instances.If two dimensions are used them the dimensions are plotted as an X-Y graph, if more than two dimensions are used then the instance space will be a three-dimensional cube and the discriminant a plane rather than a line

Support Vector Machine (SVM) : is a technique often used to increase the accuracy of a linear discriminant. A linear discriminant that includes a margin on each side of the line for classifying data. The linear discriminant line can be drawn at any angle to separate the classes, An SVM add a margin on each side of the discriminant line. It them applies mathematical equations to maximize the margins while selecting the optimal angle for the linear discriminant. the technique improves the accuracy of the linear discriminant to separate the classes. The insurer uses the optimized linear discriminant to create rule to predict.

Linear Regression : Is a type of linear function which predicts a numerical value for the target variable. It assumes that the predicted value changes proportionately (a linear relationship) with the attribute value. It is used to calculate a line that best fits in the observations in order to estimate the mean relationship between the attribute and the target variable.

The vertical distances between the dots and the line are known  are "errors" produced by the model. they are measured as the difference between the observed value of the target variable (the dot) and the mean value as estimated by the line.

The objective in calculating the regression line is to minimize total errors a common algorithm for determining the line minimizes the sum of the squared errors and is known as least squares regression. Squaring the errors increases the penalty for values further from the line compared with those closer to the line.

!!Unlike a linear discriminant, which is used to predict a class for the target variable, linear regression predicts a numerical value for the target variable as a functions of one or more attributes (explanatory variables).  Just as with a linear discriminant, linear regression can be show as a line on a two-dimensional graph.!!

Least Squares Regression : A type of linear regression that minimizes the sum of the squared vales of the errors based on the differences between the actual observed and estimated (mean) values.

Multiple Regression : A variation of linear regression whereby two or more attributes (explanatory) are used for prediction. The values of each attributes would be weighted based on its influence in estimating the value of the target variable.

Disadvantages of Linear Regression :

• the assume that the variability (as measured by the model error), or variance, around the estimated mean of the target variable is normally distributed and therefore follows a standard symmetrical pattern. But with property-casualty insurance data, variability around the mean is not normally distributed.
• Second assumption is that the variability around the estimated mean of the target variable is the same regardless of its size. This is known as homoskedasticity. But with property-casualty insurance data, variability around an estimated mean of the target variable usually varies with its size.

Homoskedasticity : In linear modeling, variance of the model errors around the estimated mean of the target variables the same regardless of its size.

Generalized Linear Models :

There are three characteristics of GLMs :

• GLMs use a link function that defines the relationship between the mean values estimated by a linear model model (called the systematic component) and the variability of the target variable around those means (called the random component).
• The random component is usually specified as an exponential distribution, which is not normally distributed. that overcomes the problem previously discussed that the random component of a linear model is normally distributed.
• The variability (variance) of an exponential distribution is a function of its mean overcoming the problem of homoskedasticity.

Explanation of GLMs : 

When estimating auto frequency, it is common to use a linear model of various explanatory variables, such as miles driven and driver experience, as the systematic component. the mean values estimated by the linear model are connected with a log link function to a Poission distribution, as exponential distribution that serves as random component. A Poisson distribution is commonly used to model a probability distribution of claims frequency.

Poisson distribution : a distribution providing the probability of a number of independent events happening in a fixed time period.

Logistic Regression : It is a classification analysis model which not only tells the classification for a new instance but also the likelihood, probability, that it falls into that class. this is known as class probability distribution, which is an estimate of the probability that an instance will fall into its predicted class.

Logistic regression is a type of generalized linear model that uses a link function based on logarithm of the odds (log odds, or logit) of the event occurring. (The odds of an event occurring is the ratio of the probability of the event occurring to the probability of the event not occurring.) 

Cluster Analysis

Is a unsupervised learning technique that can explore data to discover previously unknown information or relationships. Several iterations of cluster analysis can be applied to subdivide clusters and provide more granular information.

Hierarchical clustering

A modeling technique in which multiple clusters are grouped according to their similarities.

Dendogram : A diagram that illustrates the groupings from hierarchical clustering. The distance between clusters represents how similar each cluster is to another cluster. F clusters are grouped at one level of the hierarchy, they remain grouped as each higher level. The distance between clusters represents how similar each cluster is to another cluster.

K nearest Neighbors algorithm :

Voting Method, the prediction of the instance of the future target variable based on the majority of historic data points which are grouped as similar. the nearest neighbors that are most similar would have the greatest weight. We can make a decision on Voting Method.

There is another method used along with Voting Method, which is the average method, where the average or probability is calculated for likelihood 

K-means : An algorithm in which 'k' indicates the number of clusters and "means" represents the clusters' centroids.

Centroid : the center of a cluster is the average of the values for the instances in each cluster.

Clusters : can be described in two ways - differential and characteristic

• A characteristic description describes the typical attributes of the cluster.
• A differential description describes the differences between the instances of one cluster and those of other clusters.

Difference between K-means and Hierarchical clustering : 

• K-means clustering presented a differential description. The clustering around different characteristic points provided attributes that were different for each cluster in order to distinguish those instances.
• Hierarchical clustering is in groups according to similar attributes

Text Mining

Sources of text (unstructured data both internal and external): claims files, social media posts, news stories, and consumer reviews. 

Bot traditional and newer modeling techniques can be applied to the results of text mining.

Text is unstructured and must be cleaned up and turned into structures data before it can be used in a model. To understand the challenges and benefits of text mining, it is important to understand its process.

To apply a modeling algorithm to text, the text must be put into a form that the algorithm can recognize, such as by giving it numerical values.

Steps in text mining process:

1. Retrieve and Prepare text
1. collecting a set of writings or documents called corpus, where each document is made of various words, also called terms or tokens. (In a text mining context, a meaningful term or group of terms are called tokens)
2. Preprocessing of corpus, such as cleaning of text, removing punctuation, spaces, and numbers (depending on context) called as stopwords (are common words that offers little meaning and is filtered out in a text mining context).
3. Remove abbreviations
4. Stemming the removal of a suffix from a term for the purpose of text mining. 
2. Create a Structured Data from Unstructured Data
1. key terms can be extracted and represented in a table or spreadsheet as structured data
2. Term Frequency : a measurement of how often a term appears in a document. we generally set a upper bound and lower bound of the term frequency and analyse only those terms which fall in that inner bound.
3. Inverse document Frequency (IDF) : the measurement of the significance of a term (or rarity of a term) within a document of test in a corpus based on how many documents within which it appears in that corpus. Formula : IDF(t) = 1 + log(Total number of documents / number of documents containing it), the higher the value of IDF,  the rarer the term and higher the significance of the term and the document it is mentioned it
3. Create a model using Data Mining Techniques, 
1. To apply a modeling algorithm to text, the text must be put into a form that the algorithm can recognize, such as by giving it numerical values.
   
2. using nearest neighbors or classification learning technique, rules can be created and applied to new documents.
3. Drawback : over-fitting with too many list of words used in model, different variations of the same word.
4. After probability has been defined using classification technique, a linear scoring method provides weight to each term, based on its probability, which appears in a document. The document is then a final score, which is the probability that it indicates fraud.
4. Evaluate the Text Mining Model
1. Using a confusion matrix of the text mining results, the metrics can be calculated. a matrix that shows the predicted and actual results of a model.

Sentiment Analysis : is a way of measuring the opinion of something (often a product), be it positive, negative, or neutral. In sentiment analysis, each term is given a positive, negative or neutral value. The values are added to produce a sentiment score.

This type of analysis is known as sentiment analysis.  Sentiment analysis is a way of measuring the opinion or emotion of something be it positive, negative, or neutral in unstructured social data 

Semantic Network : a network that shows the logical relationships between concepts. In some text mining situations, particularly those involving large amounts of data and limited amounts of time-it can be beneficial to skip some of the pre-processing steps and have the computer itself find the meaningful words in unstructured data. One approach is to create a set of linguistic rules that can automatically process text, the results are represented as a semantic network. A second approach to text mining unstructured data is the use of neural networks. Neural network attempt to replicate the thought process of the human brain; however they have limitations in linguistic settings.

Social Network Analysis (or Network analysis)

The study of the nodes (vertices) and edges(lines) in a network. or in business words the links and influence of individuals and others.

These methods offer ways to observe social network connections:

• Link Analysis
• social network metrics
• Network classification

It helps in study rather the attributes of the nodes or points of information but how those data points relate to each other in the network.

"Claim Adjusters using social network analysis to trace links among the various individuals."

A social network, such as the email communication among co-workers in an office can be viewed as a sociogram. A graphic representation of a social network where each node (vertex or point in a network) is an actor and each edge (link) has a relationship.

Some graphs use arrows to distinguish that some connections flowing in only one direction, and some use weighted edges( thicker or thinner lines) to show the strength or frequency of the connections. A directed tie is an edge within a network graph that has direction, represented by an arrow, may or may not be reciprocated.

A sociogram of a large network can quickly become difficult to follow, and the best way to analyze a larger social network is through a matrix which is the representation of data through rows and columns of a social network analysis.

Small social network - Sociogram

Large Social Network - Matrix

Data mining techniques can be used to analyse there networks, both supervised and unsupervised learning techniques such as clustering can be applied.

Sentiment Analysis : is a way of measuring the opinion or emotion of something or behind a selection of text.

Link Analysis : is a key concept of network analysis. Link prediction is the process of trying to predict a pair of links. Measuring similarity, the basis of link prediction, allows insurance and risk management professionals to analyze groups of customers and/or potential customers and observe trends. "Similarity is the basis of link prediction."

Insurers must be careful in the use of social media, for privacy concerns.

Social Network Metrics

The term "path" is used to describe how something flows through a network, passing each node and edge only once. The efficiency of the flow between social network connections can be determined through centrality measures

centrality measures In  a social network context the quantification of a node's relationship to to other nodes in the same network

Degree : a measure of the connections each node has.

Closeness : the measure of the average distance, or path length, between a particular node and the other node in a network.

Betweenness : The measure of how many times a particular node is part of the shortest path between two other nodes in a network. The node serves as a connection between otherwise separated nodes in the network

Network classification : Nearest neighbor and data mining algorithms, such as logistic regression, can also be used to analyze social network.

Egonet : A network composed of an ego (personality or trait of an individual)  and the nodes (similar people) directly connected to it.

Bigraphs : is a network that has two types of nodes. This can provide a complete picture of a network

the tendency of people to connect to others who are similar to then is called homophily

Local Variable : In a social network analysis an attribute that relates to the characteristics of a specific node.

Network Variable : In  a social network analysis, an attribute that relates to the node's connection within the network.

the above two variable are used in a modeling context where each node is viewed as an instance, describes by its local and network variables. This format allows the data gleaned from social network analysis to be used in various data mining algorithm, such as classification, clustering and logistic regression.

Neural Network

Most important ability is to analyze large quantities of data. Can only operate on numerical data.

!!IMPORTANT - In order for neural networks to predict the success of a project, the factors that caused the success or failure of previous projects must be understood. !!!

Predictive Analytics : Statistical and analytical techniques used to develop models that predict future events or behaviors.

Hidden Layer : A layer in a neural network in which neurons learn and recode information received from input data. the hidden layer performs various calculation using mathematical functions to match input and output. These functions may employ cluster analysis, classification analysis, or linear functions. A hidden layer is made up of neurons which is mathematical function ins a neural network that receives input. The hidden layer of a neural network performs various mathematical functions to match inputs to outputs.

Classification analysis : a supervised learning technique to segment data according to the values of known attributes to determine the value of a categorical target variable.

Application of Neural Network

1. customer retention after premium increase
2. renewal after changes to premium
3. produce more accurate rates tailored to the individual customers
4. can pick up small nuances, that means small patterns in data
5. find factors using machine learning and cluster analysis to determine probability of failure or success of a project
6. It develops rules to make predictions as it performs various mathematical functions.
   

Drawbacks of Neural Network

1. hidden layer is too opaque, optimization of the hidden layer(s) must be precisely determined
2. Network must be able to learn from errors, however there is a risk of over fitting
3. It is more difficult to identify over-fitting and correct it
4. it cannot work on non-numerical data

Optimization in a neural network is achieved when the prediction at the output layer is evaluated to actual values. work of a data scientist to precisely determine the optimization function.

Data scientists can use clustering with neural networks to produce more accurate results. Unlike other types of data analysis techniques, neural networks can make observations.

Chapter 4 - Underwriting Applications of Big Data Analytics

Generally automobile rates are based on a class basis. that is, similar exposures are grouped and the rates of the group are charged to each member of that group. Underwriters and Actuaries identify attributes and additional attributes that reflect potential frequency and severity of loss.

The use of telematics can help gather information (additional attributes such as accelaration of car and hard braking) so that underwriter can better understand personal and commercial policyholders and to develop more accurate automobile insurance rates or it can be used as simple as a risk management tool.

It is difficult to rely solely on traditional underwriting guidelines for newly introduced products and emerging technologies, therefore advance data analytics is important

Telematics can be used through temporary or permanent installation of tracking devices, through embedded telematics equipment and through smartphone applications. these devices track driving habits, regarding braking, acceleration, speed, cornering, lane shifting, left turn versus right turns, and the time of the day the vehicle is driven.

privacy and regulatory considerations for vehicle telematics, vehicle telematics data does not typically fall under states' definitions of protected consumer information.

Purpose of Telematics devices is to track driving habits and then transmit that information wirelessly to a computer.

Traditional Attributes Rating Attributes

• Territory - which include road conditions, state safety laws, and the extent of traffic regulation are territorial factors
• Use of the Auto
• Age
• Gender
• Marital Status
• Vehicle Weight and Type
• Vehicle Use : heavy vehicles more likely to cause severe damage
• Radius of Operation : long distance increases the changes of accident
• Special Industry Classifications : food delivery, waste disposal, farmers
• Driving Record
• Driver Education
• Years of Driving Experience
• Credit Score

Usage Based Insurance :  A type of auto insurance in which the premium is based on the policyholder's driving behavior.

Advantages of Usage Base Insurance :

• Allows insurers to better segment drivers in rating classifications between preferred, standard and non-standard

Disadvantage of Usage Base Insurance :

• People who participate are likely already safe drivers.
• Another challenge is the need to evaluate the data in context so that the data is not painting the wrong picture. For example, a driver in a congested area will brake harder and more frequently than a driver in a rural area.

Loss Ratio : A ratio that measures loss and loss adjustment expenses against earned premiums and that reflects the percentage of premiums being consumed by losses. 

By using telematics loss ratios will decrease and retention ratios will increase

Telematics can be also used at risk-management level by organizations. 

Vehicle telematics can function as ongoing driver education for those who participate.

It takes sophisticated modelling techniques to make that information relevant and to determine how much the information should affect rates.

Insurers use loss exposure data generated through telematics to supplement the data they have traditionally used for rate-making. thy analyze the data using sophisticated algorithms, such as generalized linear models, to determine the correlations and interactions among all the rating attributes being considered. these includes traditional rating attributes, such as vehicle use and type, and new attributes generated through telematics, such as braking, acceleration, and time of day. Increasingly insurers are also using machine learning to aid in the discovery of variable interactions that may not be evident when using a predicted linear model

One of the areas where the telematics information will be distorted by the fact that older autos have weaker brakes and are owned by a higher proportion of younger drivers that are newer autos. Therefore younger drivers tend to experience more accidents than older drivers. There a machine learning algorithm can analyse and pick these interactions among these rating attributes.

Privacy and Regulatory Considerations for Vehicle Telematics

• Customers who do not opt to use UBI have privacy concerns regarding how their personal information is protected when it is transmitted and who owns the data about a driver's behavior.
• Vehicle telematics data does not typically fall under states' definitions of protected consumer information.
• Insurers must also ensure that their use of vehicle telematics to make rate changes is transparent and not discriminatory.

Segmenting Home Owner Policies

Historically Insurers, have struggled to make a profit in selling home owner's policies, By using more data, new correlations between granular data between their customers and loss exposures.

Before advances in machine learning and predictive analytics, homeowners policy underwriters were Limited in their ability to incorporate the many attributes needed to accurately predict losses.

To develop the sophisticated model needed to refine its homeowners classifications, the insurer must choose the criteria it will use to segment its homeowner policies.

Traditional rating variables :

• year built
• construction type
• location
• age of home electrical's system (additional attribute for risk grouping)

Geo-location information (can be obtained from governmental wildfire agency) :

• underbrush
• relative wind speeds
• distance between home and fuel could help to better understand the increasing wild-fire related claim

Customers with a lower loss ratio are more profitable. Using analytics like machine learning, the insurer has managed to achieve its goal, it has increased rates, lowered overall ratios, and become more competitive in the marketplace.

Using traditional flat rate increase across the homeowners book, the customers with higher loss ratios would be more likely to stay, leaving the insurer with the least profitable segment of its book.  The overall loss ratio would most likely not improve.

Personal lines insurers are investigating ways to incorporate discounts for homeowners insurance policies based on homeowners' use of smart home devices. Personal lines insurers may even develop usage-based homeowners insurance, similar to that now offered for auto insurance, in which rates would be partially based on wireless data sent from devices within the home.

Relative wind speed is an attribute that could help an insurer better understand wildfire-related claims.

By using machine learning to segment its policies, the insurer see the effect of attributes and interactions it had not previously considered in its traditional rating model. The insurer notices that loss ratios are inconsistent across the segments based on the interactions between traditional and new variables. Producers' and claim representatives' files may contain a significant amount of data that could be made available to machine learning.

Customers with a lower loss ratios - who are therefore more profitable to insurer - seek a competitive rate; the challenge for an insurer is to provide this while still challenging sufficient premiums for customers with higher loss ratios, who are unlikely to leave.

Without machine learning , finding variables would be really time-consuming

Underwriting Products Liability Risks using Data Mining

The data science team will apply cluster analysis, as unsupervised learning technique to find out whether there are any patterns in the recent large claims sometimes called as K-Means or nearest neighbor clustering. Supervised learning and predictive modelling require known attributes and a known target variable.

Use social media data and use text mining to search social media for references to the products manufactured by the insurer's customers. These can be equated to a positive sentiment or a negative sentiment which will give a sentiment score to the product.

Using this information along with insurance professionals knowledge, the data science team has the attributes and the target variable with which to build a predictive model.

• Predicting Liability risks is important, because retailers, manufacturers, distributors can become liable when products cause injuries to consumers. 
• Social philosophy favors consumers, 
• and laws that hold businesses liable for their products have increasingly expanded.

Data Mining can help an insurer properly evaluate an account's products liability loss exposures. Particularly when evaluating products with uncertain risk characteristics, underwriters can benefit for this approach, rather than rely on traditional underwriting guidelines and also find out new trends in risks which they may not have factored it in during underwriting.

Predictive Modeling for Emerging Risks

The insurer will also use the predictive model to underwrite new accounts. by more accurately predicting the likelihood of claims, the insurer will be able to determine which accounts it should accept to insure and what pricing levels are required to maintain profitability for the line of business.

Monitoring reactions to products and its liability can be monitored through text mining of social media. and it can also mine its historic liability claims to more accurately price similar exposures.

Through text mining, insurers can identify attributes and the target variable with which to build a predictive model.

In a more realistic and complex scenario, the insurer, working with the data scientists to analyse the data mining and using its insurance professionals' knowledge and past experience with liability claims, could determine which one of the products' attributes are important in terms of product liability.

In terms of risk management perspective, the insurer can use the valuable information it has gained from data mining to help its customers.

Machine learning allows insurers to incorporate more attributes in their underwriting decisions. A fully trained segmentation model can find meaningful patterns in data and automatically segment policies. 

Data Mining can help an insurer properly evaluate an account's products liability loss exposures. 

Cluster Analysis , text mining of social media and predictive modelling can help to predict claims, and therefore, an account;s loss exposures.

Chapter 5 - Claims Applications of Big Data Analytics

Mainly used in Fraud Detection and claim adjuster assignment for claims predicted as severe, about 10 percent of the property-casualty insurance industry's incurred losses and loss adjustment expenses stem from fraud. Insurer's attempt to detect fraud by identifying patterns, controlling underwriting at the points of sale and using special investigation unit (SLU)

Advances in data mining to more effectively identify patterns in fraudulent  claims activity. Insurance and Risk Management professionals therefore benefit from understanding how to analyse links in a social network and clusters of data points to capture new trends in claims fraud.

Claims representatives can often find evidence that someone may be lying by comparing his or her social media posts with his or her statements in a claim

Insurer's data science team mainly use cluster analysis and classification trees.

Cluster analysis can be used to create cluster of similar claims according to various attributes. These attributes are not known in advance because cluster analysis is an explanatory technique without any predefined variables. Clusters of claims instance would be expected to group around attributes associated with claim severity. Cluster analysis would continue to be applied on individual clusters until the data scientist has a good understanding of the relationships between significant variables.

After cluster analysis is complete, the characteristics of the clusters can be used as attributes to develop classification trees with which to assign new claims.

Logistic Regression could then be used to estimate the probability of each of these outcomes to enable the insurer to determine the appropriate resources for each complex claim.

Predictive modeling seems to be the long term solution to detect fraud claims, without increasing auto rates and spend on resources to investigate claim of fraud in nature

Steps taken to identify fraudulent claims :

1. Detect claims fraud through traditional fraud indicators (lists of claims fraud indicators are published by National Insurance Crime Bureau) plus insurance company can make their own list of fraud indicators and also through mining social media data.
2. Apply network analysis by examining links and suspicious connections.
3. Apply cluster analysis to discover claims characteristics that might indicate fraud.

Examples of fraud indicators an insured or a claimant pushes for quick settlement or has too much or too little documentation

Although the insurer is taking steps to identify fraudulent claims, some fraud still goes undetected, the reason is that the traditional fraud indicators is based on historical data. Intelligent and innovative fraudsters will change their approaches and patterns, limiting the usefulness of these indicators. The traditional approach is highly subjective and depends on claims representatives' experience in the field. A more automated approach would allow for greater objectivity and enable new claims representatives to be more effective in less time.

In analyzing social media to detect fraud claims practices, requires not only investigating social media posts but also the connections within a network as well. The connections in a network would help in identifying a fraud ring

Because for predictive modelling an insurance may have very less historical data to indicate fraud claim and since fraud is ever evolving, the model may become quickly outdated. Clustering techniques such as K-means are unsupervised learning techniques to identify new fraud indicators before predictive modeling is applied. Essentially the fraudulent claims are outliers within the already outlying cluster

Using classification tree analysis  in claims assignment :

In classification analysis, A computer recursively applies a model to analyze different splits in the values of attributes.

A relatively small percentage of claims account for much of the costs to insurers. some of the high-cost claims can be easily identified at the time of first report, such as the total loss of a building from fire, a major auto accident, or a plant explosion. However many claims develop gradually into significant losses. 

Lift is the percentage of a model's positive predictions divided by the percentage expected by chance.

Example Workers Compensation Claims, where the majority or more than half costs to be paid are medical costs.  Potentially complex claims are the most difficult for insurers to identify at the time of first report.

To use the resources effectively, Insurers or Risk Managers can use their own data, or they an use their insurers' or their parties' data to identify potentially serious claim.

Interest in Modeling is also used to determine claimant characteristics to determine if the person will likely have more chronic problem or develop opoids problem.

Classification tree analysis can be used to assign new claims according to target variables.

Complex Claim : A claim that contains one or more characteristics that cause it to cost more than the average claim.

Procedure to create a model for complex claims

1. An important step after identifying the attributes of complex claims is ranking the attributes according to their relative information gain.
   
2. Determine list of attributes (input variables) and ranked  by the relative importance using Information Gain by complex claims attributes. Including a claims professional on the data science team can help with selection of the most important attributes.
3. Using a classification tree to illustrate attributes of complex claims to build a predictive model. The computer recursively built a tree by analyzing different splits in the values of attributes in various sequences.
4. Validating the complex chain model. Through the machine learning, the model adjusts the weights assigned to each of the attributes to better predict accuracy. This is done to keep the model effective based on new incoming data.

!! Note : The sequence of attributes in the tree does not strictly follow the relative information-gain ranking for the attributes was developed independently of other attributes. that is usually the case, because the information-gain ranking for the attributes was developed independently of the other attributes. However, the attributes Representative by each node in the tree depend on the attributes (and their values) that sit above them !!

Combination of Nodes : A representation of a data attribute in a classification tree.

Complex Claim Reporting : 

1. Claim reported online or by telephone
2. Direct the claim to the appropriate claim intake based on geographical location
3. If identified as catastrophic claims, it is assigned to senior claims adjuster
4. Machine learning algorithm is used to model and predict the type of claim as "complex" or "not complex"
1. It should be decided based on attributes on what type of modelling technique to use
5. through machine learning, the model adjusts weights assigned to each of the attributes to better predict complexity.

Claims payment are the biggest financial obligations for the insurer

To illustrate how to improve claims processes using data analytics

1. Diagram the claims handling process, there are six major activities in the claims handling process:

• acknowledging and assigning the claims
• identifying the policy
• contacting the insured or the insured's representative
• investigating and documenting the claim
• determining the cause of loss, liability, and the loss amount
• concluding the claim
• identify how the loss occurred, witness, potential fraud or subrogation opportunities, and the loss amount

2. Decide how business process analytics can be applied to this process

• to improve customer service, efficiency, and cost-effectiveness
• BPM (Business Process Management) : A systemic, iterative plan to analyze and improve business processes through life-cycle phases to achieve long-term goals and client satisfaction
• Data analysis is used in gathering intelligence as well as developing models

1. 
2. Process Mining applied to claims

• starts with data analysis of claims activity logs
• uses data analysis technique to explore data regarding an existing process, such as claims handling, to provide an understanding of the entire process. Process mining differs from typical data mining on its application of results to an entire process rather than one specific area, such as claims costs.
• mining of the claims activity logs to identify claims handling activities and the timing of them. the first phase in the process mining of the insurer's auto claims involves process discovery of claims activities. which shows the date and time when each activity began and ended - an example indicator is Reviewed policy
• This information is then used to design classification tree models, for claims activities, such as claims assignment, fraud investigation, and subrogation potential

Process Mining : The use of exploratory data analysis to provide insights about a business process and identify potential improvements. Process is like data mining. It differs from typical data mining in its applications of results to an entire process rather than one specific area, such as claims costs

Process Discovery : The use of data analysis to identify the activities in a current process. it is sued to develop a process map of current claims activities, referred to as process instances, and the average duration of each process instance.

Process Map : A diagram or workflow of a process that is based on the results of process discovery. which allows the insurer to know the average length of time for each process instance in a process and set benchmarks, made from claims logs.The process map would help insurance set the initial benchmarks for each activity in the current claims handling process based on internal metrics.  The process map is a diagram of the process that is based on the results of the process discovery.

Process instance :  A discrete activity of a business process.

After process map, The insurer will need to combine process mining with data analysis techniques to gain a better understanding of what is involved in the investigation and resolution of different types of claims. Data analysis techniques include - classification techniques and clustering

Classification tree models can be used to design models for claims activities, such as claims assignment, fraud detection, and subrogation potential.

Business Process Management (BPM) : A systematic, iterative plan to analyse and improve business processes through life-cycle phases to achieve long-term goals and client satisfaction. The goal of BPM in claims handling is to improve claims efficiency, customer service and cost-effectiveness. BPM begins with explanatory analysis of the current process and then develops models for improvement. Data analysis is used on gathering intelligence as well as developing model.

Cluster analysis is an exploratory technique without predefined variables that can be used when attributes are not known.

An insurer should be prepared to Reevaluate the attributes in a predictive model.

To improve claims handling process, using of Cluster analysis is an exploratory technique without predefined variables that can be used when attributes are not known.

Chapter 6 - Risk Management Application of Data Analytics

In Workers Compensation Claims, through past accident data, it is often found that accident causation focuses on either single unsafe act or condition. However, even the simplest accidents  are the byproduct of multiple actions and complex interactions of conditions.

• Using sensors is a powerful way to collect data about biological condition of workers and physical conditions of workplace environment. 
• Using this data we can run advance predictive modelling or Machine Learning to improve the accuracy of forecasting accidents.

After data from sensors is categorized to develop attributes that help predict workplace accidents, these attributes are combined with traditional workplace accident attributes and then analyzed independently for their information gain.

An organization's safety program may combine elements of these accident analysis techniques. For example, Fault Tree Analysis (FTA),which incorporates RCA concepts and FTA Techniques, can be used to identify potential accidents and predict the most likely system failures. Categories of Workplace accident analysis technique include :

• System Safety : A safety engineering technique also used as an approach to accident causation that considers the mutual effects of the interrelated elements of a system on one another throughout the system's life cycle. It analysis hazards and causes of hazards and estimate the probability of particular kinds of breakdowns and suggest cost-effective ways to prevent these system failures. It examines the organizations as a whole, and relies on specific techniques for determining how these hazards can lead to a system failures and accidents.
• Root Cause Analysis (RCA) : A systematics procedure that uses the results of the other analysis techniques to identify the predominant cause of the accident. it is used to identify the root cause of the accident and mitigate future actions, inactions, conditions or behaviour of such events.
• Failure Mode and Effects analysis (FMEA) : An analysis technique the reverses the direction of reasoning by starting with causes and branching out to consequences. The goal is to prevent or reduce the severity of a harmful event and address the event based on priority. The goal is is to prevent or reduce the severity of a harmful event by identifying the order in which critical failures should be addressed. The specific actions (consequences) include eliminating the failure mode, minimizing, the severity (consequences). reducing the occurrence, and improving detection.
• Criticality Analysis : An analysis the identifies the critical components of a system and ranks the severity of losing each component. The goal is to prevent or reduce the severity of a harmful event and address the event based on priority. The goal is to prevent or reduce the severity of a harmful event by identifying the order in which critical failures should be addressed. The specific actions (consequences) include eliminating the failure mode, minimizing, the severity (consequences). reducing the occurrence, and improving detection.
• Fault Tree Analysis (FTA) : An analysis that takes a particular system failure and traces the events leading to the system failure backwards in time.Incorporates RCA concepts and FMEA techniques to identify various ways of "breaking" the fault tree; that is, it interrupts the sequence of events leading to system failure so that the failure itself can be prevented.

Failure mode and effects analysis is a traditional analysis technique that, when paired with criticality analysis, helps an organization prevent or reduce the severity of a harmful event by identifying the order in which critical failures should be addressed is failure mode and effects analysis.

In Fault Tree Analysis (FTA)  the risk manager will examine the series of events and conditions that led to an accident. The accident appears at the op of the fault tree, and the events necessary to produce it appear as branches. This is known as fault tree "TOP EVENT"

• The tree's branches are connected by "and" gates and "or" gates. These gates represent the casual relationship between events, which are depicted as rectangles within the tree.
• A fault tree also can be used to calculate the probability of an accident if the probabilities of the casual events are known.

Limitations of FTA : If there is a high degree of uncertainty with underlying or base events, the probability of the accident may be ascertain. Additionally, important pathways to the accident might not be explored if some casual events are not included in the fault tree. 

!! traditional analysis such as FTA  cannot easily account for human error and conditional failures (which cause one another in succession, culminating in an accident) !!

Holter monitors to detect cardiac arrests is an example of sensors. Limitations of sensors, is that if we cannot find the relationships among accumulated data points from all the sensors, which would explain the environmental conditions, worker attributes, and other relevant factors which affect the probability of an accident occurring. Its use will be not be fully realized

Steps to do Data Analytics

1. Collect the data and categorize.
2. The categorize are used to develop into similar attributes
3. Each attribute is then analyzed independently for its information gain
4. Supervised learning technique such as Classification tree algorithm is applied to training data. The algorithm executes recursively and shows how various combinations of attributes can be used to predict the probability of an accident occurring 
5. Finally, holdout data is used to test the model and determine its predictive power on data that was not used in model development
6. Classification results from the combination of attributes connected to the rectangle through the sequence of arrows, each of which depicts the actual value of the attribute to which it is connected.
7. Probabilities are added to each of the rectangles based on the percentage of time the model correctly predicts the specified class when applied to the training data.

The Goal of the Classification tree algorithm is to determine the probability of an accident occurring or not occurring.

Assessing Reputation Risk through text mining and social network analysis

Reputation is an intangible asset the relates to an organization's goals and values, results from the behaviors and opinions of its stakeholders (stakeholder perception) and grows over time. It involves a comparison of stakeholder's experience and their expectations and is the pillar of the organization's legitimacy, or social license to operate.

Data analytics offers organization a way to keep up to date on the shifting perceptions regarding their services and products. Text mining and social network analysis will be used to come up with a plan of action.

An organization's risk manager can better target a response to a crisis by considering the combination of local variables and network variables in a social network.

Reputation Risk : The risk the negative publicity, whether true or not will damage a company's reputation and its ability to operate its business, it involves managing the organization's interactions with the general public, its stakeholders an its employees. Risk managers must use a systemic approach, to carefully manage the organizations interactions with the general public, its stakeholers and its employees.

Steps in Text Mining :

• find attributes to perform sentiment analysis one
• clean up by elimination spaces, punctuation's, and words that provide little information (stop words)
• reducing and words to their stems by removing prefixes and suffixes (stemming)
• create two dictionaries one with favorable words and another with unfavorable words
• Perform sentiment analysis term frequency (TF) and inverse document frequency(IDF) of certain terms

For example : One term is used multiple times within only one of the blog posts. It has a high TF within blog post and high IDF, which is a measure of its rarity within the corpus of blogs. The combination of high TF within the blog post and high IDF indicates that the term is worthy of consideration.

Social Network Analysis

• Sociogram is for a small social network
• Matrix is for very large social network

Data scientists will examine social network connections through a sociogram to understand how far a negative sentiment that originates with one person or a few people can spread. The influence of the negative sentiment significantly depends on its posters' metrics given below:-

• Closeness: The measure of the average distance, or path length, between a particular node and the other node in a network
• Degree: The measure of the number of connections each node has
• Betweenness: The measure of how many times a particular node is part of the shortest path between two other nodes in a network.

!! Using Text Mining and Social Network Analysis, targeted messages can be sent to the correct group with more influence and effectively respond to reputation risk. !!

The centrality measures of degree, closeness, and betweenness matter not only in the context of one network, but also in the context of how one person or a few people can connect multiple social media networks.

An organization's risk manager can better target a response to a crisis by considering the combination of local variables and network variables in a social network.

A sociogram is particularly useful for analyzing small social networks

!! Organization should take precautions not to specifically identify individual customers, by combining pieces of information of the user. This in itself is a reputation risk . Data scientists continue to work on developing new ways to ensure individuals' privacy while still allowing organizations to reap the benefits of network analysis.!!

Performing a network analysis after performing text mining on social data can provide information about how far and how quickly a sentiment has and could spread.

Using Clustering and Linear Modelling for Loss Development

It is important for risk management professionals to determine the ultimate value of losses as accurately as possible, as this is the biggest obligation of the insurer, the money left can be invested for insurance operations.

Long-Tail Claim : A claim in which there is a duration of more than one year from the date of loss to closure. These claims are difficult to predict due to multiple factors.

Ultimate Loss : The final paid amount for all losses in an accident year.

Adverse Development : Increasing claims costs for which the reserves are inadequate. If factors of adverse is not known then unsupervised learning like K-Means clustering can be sued to find patterns. Examples of Outliers can be Claim Size and Ratio of Incurred Loss to incurred loss at the eighteen month evaluation point (ratio > 3,5)

Excess Liability Insurance : Insurance coverage for losses that exceed the limits of underlying insurance coverage or a retention amount.

Cluster analysis is an exploratory technique without predefined variables that can be used when attributes are not known.: The sum of the paid losses and loss reserves and loss adjustment expense reserves. General liability losses are typically assumed to reach their ultimate loss value within 66 months.

Loss Development factors, usually based on previous accident year's loss development data, are multiplied by the total incurred losses for each accident year to reach an estimate of the ultimate loss for that year.

"A closed claim is assumed to have reached the ultimate loss value."

Typical attributes of high severity claims

• Respond more than one week after the date of accident.
• Liability denied by the claim representative.
• Claimant represented by an attorney.
• Large Jury Awards

With the above factors of high severity claims, Data scientist can use a Generalized Linear Model (GLM) to project the average ultimate severity as the target variable for these types of claims based on their attributes. The claims reserves should be based on the results of the GLM.

Predictions of ultimate losses for specific accident years are based on total incurred losses. After a predictive model is developed to evaluate estimates of ultimate losses, the analysis should be repeated to determine accuracy.

Summary :

Workplace accidents and their causes are ideal subjects for the application of data analytics because traditional methods of workplace accidents analysis rely on human interpretation of limited amounts of information and cannot always account for how contributing factors combine to cause an accident.

Chapter 7 - Implementing a Data Analytics Strategy

In order to implement a sound and successful data analytics strategy - SWOT analysis is needed to be done. Especially  for P&C  insurers when the change entails a fundamental shift in philosophy, such as the one from reliance on traditional organizational infrastructures to data-driven analysis. The P &C Insurance industry, rooted in risk aversion and predicated on long-term strategy, has traditionally been low to embrace change.

The emergence of data analytics-can therefore potentially gain a marketplace edge by adopting new practices or efficiently than its competitors. An organization's integration of data analytics begins with the acceptance of the fundamental concept that more decision-making information, used intelligently, is always better. One way an insurer can determine this is through SWOT analysis.

Underwriters can effectively analyze policy segmentation to make more-informed rate changes and improve loss ratios and customer retention when using a fully trained segmentation model through data analytics to find meaningful patterns in data and automatically segment policies.

An insurer could use predictive modeling to prioritize a claim for investigation based on the probability that it is fraudulent. Data mining techniques, such as text mining, social network analysis, and cluster analysis, can be used to extract the data.

SWOT (strength, weakness, opportunities and threats) Analysis : A method of evaluating the internal and external environments by assessing an organization's internal strengths and weakness and its external opportunities and threats. It allows an organization to consider both the general environment and its immediate environment. The method was devised by Albert S. Humphrey, a business scholar and management consultant who specialized in business planning and change. The approach used by organizations varies based on each company's needs. In SWOT Analysis a company should also thoroughly analyze how they affect its strategic plan.

Because the SWOT list can be extensive , Understanding the business issue that has prompted the SWOT evaluation is very important. In the case of potential use of Data Analytics enables use of management to develop two general perspectives:

• Internal (organizational) strengths and weakness can be identified as financial, physical, human and organizational assets. based on factors like managerial expertise, available product lines, staff competencies, current strategies, customer loyalty, growth levels, organizational structure and distribution channels organizational structure and distribution channels. In a data analytics project, the risk manager's consideration of the internal environment begins with clarifying the objective of the project.
• External factors, which can be identified through trend analysis, it can be presented by new opportunities in new markets, possible acquisition targets, or a reduction in competition, economic downturns, or changes in customer preferences.

A SWOT analysis focused specifically on the potential use of data analytics enables management to identify internal (organizational) strengths and weaknesses that could potentially be affected by a data analytics initiative and external factors that present opportunities for growth or threats to the organization's survival.

SWOT Analysis Process

Is done through a group activity that involves an organization's managers and is organized by a facilitator

• Brainstorming 
• If the group is large, the facilitator may divide the managers into smaller groups to encourage participation
• through brainstorming, factors are listed under each of the SWOT headings.
• This activity will produce many factors organized under each SWOT heading.
• Refining
• To make the list easier to examine, similar items are clustered together.
• Items of high importance are noted under each of the SWOT headings.
• Prioritizing
• Strengths are ordered by quality and relative importance.
• Weakness are ordered by the degree to which they affect performance and by their relative importance.
• Opportunities are ordered by degree and probability of success.
• Threats are ordered by degree and probability of occurrence.
• Strengths that require little or no operational changes and can be paired with opportunities are designated for potential action to maximize competitive advantages that will entail low risk
• Weaknesses that can be paired with threats are designated in the prioritized list for potential action to minimize consequences that entail high risk

Through Data Analysis a business can do below three

• Reinforce Strengths by increasing its profitable business, although some opportunities may lie in reducing in costs as well.

• Mitigate Weakness by using predictive modelling techniques like social network analysis, text mining and cluster analysis to better identify fraudulent claims. Use Machine Learning to identify patterns in underwriter guidelines, loss ratio to improve policy segmentation methods. Help in data driven decision making by intelligently deployed information
• Inability to identify fraudulent claims
• Diminishing effectiveness of existing policy segmentation methods
• Pervasive conventional thinking

• Exploit Opportunities by hiring better talent using predictive modelling, analyze employee turnover and incentivize employees to remain engaged. Use text mining and social network analysis to accumulate as much publicly available intelligence about competitors. Make use of sentiment analysis algorithms, social network analysis, and other modelling techniques  to measure the public attitudes about companies insurance policies.
• Scarcity of new workers willing to enter the insurance industry : Recruitment and retention challenges posed by the scarce entrance of new workers, customers' increased access to competitive information, and cynical public attitudes about insurers that create a more permissive environment for claims fraud are some of the technological, sociological, and industry trends that could fundamentally change the way insurers do business.
• Customers' increased access to information about competitors' prices
• Cynical public attitudes about insurers : An insurer can use sentiment analysis algorithms, social network analysis, and other means of measuring public views contributing to claims fraud to address cynical public attitudes about insurers.

Implementation of data analytics project and Management of inherent risks to implement a data analytics project is very important as the initiative itself

One way to manage risk is through project management. Project Risk Management aims to optimize risk levels to achieve the project goals. Its underlying structured process allows risk managers to identify and assess a project's risks and to respond appropriately.

In a data analytics project, the risk manager's consideration of the internal environment begins with clarifying the objective of the project.

Another way to manage risks is through Enterprise Risk Management (ERM) model where activities is done in five steps

• Scan Environment : Can be accomplished with an analysis of an organization's internal and external environments (external include Technological, Legal & regulatory and Political). In scanning the environment, the risk manger for an organization's data analytics initiative first needs to develop a sense of how the initiative will merge with its existing infrastructure and how the project could affect the organization's stakeholders. To hone the project's goal, the risk manager could interview various internal stakeholders and develop a series of quantifiable objectives, that, together constitute the project's overall goal.In a data analytics project, the risk manager's consideration of the internal environment begins with clarifying the objective of the project.
• The technological aspect of a data analytics initiative that involves the use of electronic monitoring and reporting equipment that must be installed in remote locations and tested and maintained for accurate reporting, security, proper location, and dependability is part of scanning the external environment.
  
• Identify Risks : This is to determine which potential risks will require treatment.
• Analyse Risks : It is done to identify the components activities of the project and then examine (by judging the likelihood and severity) the risk for each activity. The risk manager can the prioritize the risk based on the highest priority given to risk which will extend the project beyond its timelines.
• Treat Risks : Is done by avoidance, modification, transfer , retention or exploitation. However the opportunities that accompany the risk should not be overlooked. In addition project managers will involve contingency planning, which involves establishing alternative procedures for possible events thereby keeping the entire project on track within the project constraints.
• Monitor and Assure : To ensure each activity is within variance acceptable limits and in the budget allocation. Monitoring and assuring the data analytics project as it progresses involves the project manager comparing the quality, time, budget, and other constraints established for the project with the project status and determining whether the project goal will be achieved or whether resources should be reallocated to achieve the goal.

ERM is done to ensure the risk management is in sync with the strategic goals and operational objectives of the specific project. This project envision real time data to make decision on the fly.

Risk avoidance, modification, transfer, retention, or exploitation, as well as contingency planning, are some of the major options for treating risk and keeping the entire project on track within variance requirements.

Data analytics change Management

The change management process entails is particularly difficult for a company when the change entails a fundamental shift in philosophy, such as the one from reliance on traditional organizational infrastructures to data-driven analysis.

An insurer's data analytics initiative can succeed only if the entire organization understands the value of intelligently deployed information and how to apply data-driven decision making.

Alignment of the data analytics project's goals with upper management's long-term organizational objectives is an example of component activities associated with the data analytics initiative.

An organization's integration of data analytics begins with its acceptance of the fundamental concept that, when used intelligently, more decision-making information is always better.

To effectively articulate throughout the organization the need for a shift to data-driven processes as part of a data analytics initiative, top management should frame it as being urgent and crucial to the insurer's ability to acquire and retain business because competitors use data analytics techniques to improve their underwriting, pricing, fraud detection, and marketing.

An insurer's data analytics initiative should be guided by a vision statement that succinctly aligns, directs, and inspires collective action toward a common end point that improves the organization's measurable results.

The costs to implement machine learning techniques and to recruit, train, and retain personnel capable of executing data analytics strategies are costs related to the investment required to launch a data analytics technique that could undermine an insurer's cost-reduction strategy.

The leadership team for a major transformation in an organization is usually made up of one highly visible change agent and a collaborative team of sponsors of the change.

Analyzing the risks associated with the data analytics project includes establishing time estimates for crucial activities and assigning acceptable variances to be referenced in the event that they could cause results to fall outside the acceptable parameters.

• Articulate the need for change : Sense of Urgency helps persuade key individuals to invest in the change and creates the momentum helps persuade key individuals to invest in the change and creates the momentum required to spur the organization to action. 

  Establishing a sense of urgency helps managers persuade key individuals to invest in the change and creates the momentum required to spur the organization to action.

• Appoint a leadership team :  team of carefully selected individuals who recognize the need for change and provide solid, broad-based support for accomplishing it and create a Center of Excellence (COE)
• Develop a written statement of the vision and strategies : that is realistic, desirable, feasible, motivational, and focused enough to guide decision making. A common error is to mistake plans and programs for an appropriate vision.
• It provides a general direction for improving the organization's products, services, cost control, and/or relationships with customers and stakeholders
• It can be clearly described in five minutes or less
• It considers trends in technology and the market
• It is stated in positive terms
• Communicate the vision and strategies : can be in the form or arguments and eliminating signals against the message. the overarching vision and the mail supporting strategies to all employers using multiple fourms to ensure that everyone sees and hears the message repeatedly and consistently
• Eliminate the barriers of change : in the form of technological limitations, timelines, skill level and organization culture. The leadership team can try to eliminate barriers to change by establishing timelines that are reasonable and not inappropriately ambitious.
• Recognize incremental successes : Identify milestones through published goals and team charters. split goals into smaller goals this ensures the team communicate the relationship and relevance of the assigned actions to the overarching goal. Management and the COE may use milestones in a data analytics project to represent visible improvements or short-term accomplishments that are part of the progress toward the project goal.
• Entrench the change : New systems, process and structures should be created so that employees can get the benefit of the new system in place and the positive outcome. Shared attitudes, values, goals, and persistent norms of behavior must be aligned with the change to data-driven philosophies.

A Center of Excellence (COE) address the need of strong business leadership and data analytics leadership. A COE consists of personnel dedicated solely to developing data analytics strategies, formulating department-specific data analytic goals, implementing plans to encourage organization-wide collaboration on data analytics initiatives, and overseeing the evolution of a data analytics model from concept to adoption.

The members of COE are responsible for creating the vision of the change and selling it to others. also members of the COE should have the power to eliminate any obstacles to progress, expertise relevant to the change, the credibility to convince others that the need for change should be taken seriously, and leadership to catalyse the change process.

Because it is neither practical nor feasible for a risk manager to identify all of the risks associated with a data analytics project, it is important to identify key and emerging risks so they can be analyzed for their effect on the project.

Upon completion of the data analytics project, individuals may not be able to see the broader perspective or effect on the change while it is taking place, so management may need to reinforce the vision.

Risk avoidance, modification, transfer, retention, or exploitation, as well as contingency planning, are some of the major options for treating risk and keeping the entire project on track within variance requirements.