The Central Idea
Finance, insurance, and risk problems share one fundamental structure:
\[ \boxed{\text{future outcomes are uncertain, but their probabilities and consequences can be modeled}} \]A model usually identifies:
- possible outcomes;
- their probabilities;
- the gain or loss associated with each outcome;
- a rule for comparing alternatives.
Expected value is often the first summary, but risk analysis usually requires more than expected value alone.
Random Financial Outcomes
Let \(X\) represent a random financial outcome.
Possible values may include:
\[ x_1,x_2,\ldots,x_n \]with probabilities
\[ p_1,p_2,\ldots,p_n. \]The probabilities satisfy
\[ \boxed{\sum_{i=1}^{n}p_i=1}. \]Expected Value
The expected value of a discrete random outcome is
\[ \boxed{ E[X]=\sum_i p_ix_i }. \]Expected value is a probability-weighted average.
It does not say which outcome will occur in one particular case.
A First Financial Example
Suppose an investment has:
- a \(60\%\) chance of gaining \(\$200\);
- a \(40\%\) chance of losing \(\$100\).
Let gains be positive and losses negative.
\[ E[X] = 0.60(200)+0.40(-100). \]Therefore,
\[ E[X] = 120-40 = \boxed{\$80}. \]Expected Value Is Not Guaranteed Profit
An expected value of
\[ \$80 \]does not mean the investment will earn \(\$80\).
In this example, the actual one-period outcomes are
\[ \$200 \]or
\[ -\$100. \]Expected value summarizes repeated uncertainty.
Returns
A financial return measures change relative to an initial value.
If an asset begins at value \(V_0\) and ends at \(V_1\), the simple return is
\[ \boxed{ R=\frac{V_1-V_0}{V_0} }. \]If a \(\$100\) investment grows to \(\$108\),
\[ R=\frac{108-100}{100}=\boxed{0.08=8\%}. \]Random Returns
If future asset value is uncertain, return becomes a random variable.
Suppose:
| Scenario | Probability | Return |
|---|---|---|
| Strong | 0.30 | 20% |
| Normal | 0.50 | 8% |
| Weak | 0.20 | -10% |
Expected return is
\[ E[R] = 0.30(0.20)+0.50(0.08)+0.20(-0.10). \]Thus
\[ E[R] = 0.06+0.04-0.02 = \boxed{0.08=8\%}. \]Risk as Variability
Two investments can have the same expected return but very different uncertainty.
Variance measures spread around the mean:
\[ \boxed{ \operatorname{Var}(X) = E[(X-\mu)^2] }. \]For a discrete variable,
\[ \operatorname{Var}(X) = \sum_i p_i(x_i-\mu)^2. \]Standard Deviation
Standard deviation is
\[ \boxed{ \sigma=\sqrt{\operatorname{Var}(X)} }. \]Unlike variance, standard deviation has the same units as the original variable.
In return models, it is commonly used as one measure of volatility.
Same Mean, Different Risk
Consider two investments:
Investment A always returns
\[ 5\%. \]Investment B returns
\[ 15\% \]half the time and
\[ -5\% \]half the time.
Both have expected return
\[ \boxed{5\%}. \]But B has greater variability.
Loss Variables
Insurance often models a nonnegative random loss \(L\).
For example,
\[ L= \begin{cases} 0,&\text{no claim},\\ 1000,&\text{small claim},\\ 10000,&\text{large claim}. \end{cases} \]The expected loss is
\[ \boxed{E[L]=\sum_i p_i\ell_i}. \]Expected Insurance Loss
Suppose a policy has:
- \(90\%\) probability of no claim;
- \(8\%\) probability of a \(\$1{,}000\) claim;
- \(2\%\) probability of a \(\$10{,}000\) claim.
Then
\[ E[L] = 0.90(0)+0.08(1000)+0.02(10000). \]Therefore,
\[ E[L] = 80+200 = \boxed{\$280}. \]Pure Premium
In a simplified insurance model, the pure premium is the expected claim cost:
\[ \boxed{\text{pure premium}=E[L]}. \]Real insurance premiums may also need to reflect:
- administrative expenses;
- capital costs;
- risk margins;
- taxes and fees;
- profit or contingency margins;
- policy features.
Frequency and Severity
Insurance losses are often separated into:
- frequency — how many claims occur;
- severity — how large each claim is.
If \(N\) is the number of claims and \(X_i\) is claim \(i\)'s size, aggregate loss is
\[ \boxed{ S=X_1+X_2+\cdots+X_N }. \]Compound-Loss Expectation
Under suitable independence assumptions,
\[ \boxed{ E[S]=E[N]E[X] }. \]This connects expected claim frequency with expected claim severity.
Compound-Loss Example
Suppose an insurance portfolio expects
\[ E[N]=40 \]claims per year and average claim severity
\[ E[X]=\$750. \]Expected annual aggregate loss is
\[ E[S] = 40(750) = \boxed{\$30{,}000}. \]Bernoulli Loss Model
Suppose an insured event occurs with probability \(p\) and produces fixed loss \(c\).
Let
\[ I= \begin{cases} 1,&\text{loss occurs},\\ 0,&\text{otherwise}. \end{cases} \]Then
\[ L=cI. \]Expected loss is
\[ \boxed{E[L]=cp}. \]Deductibles
A deductible makes the insured absorb the first portion of a loss.
For loss \(L\) and deductible \(d\), a common payment model is
\[ \boxed{ Y=(L-d)_+ = \max(L-d,0) }. \]Deductible Example
Suppose a claim loss is
\[ L=\$4{,}000 \]and the deductible is
\[ d=\$1{,}000. \]Then insurer payment is
\[ Y=\max(4000-1000,0) = \boxed{\$3{,}000}. \]Policy Limits
A policy may also cap the insurer's payment at a maximum amount \(u\).
A simple payment rule with deductible \(d\) and limit \(u\) is
\[ \boxed{ Y=\min\big((L-d)_+,u\big) }. \]Contract structure changes the distribution of insured loss and insurer payment.
Risk Pooling
Insurance works partly by combining many uncertain individual risks.
If losses are sufficiently independent or weakly dependent, aggregate average loss can become more predictable as the pool grows.
This is a practical consequence of the law of large numbers.
A Simple Pooling Example
Suppose each of \(n\) independent policies has expected loss
\[ \mu \]and variance
\[ \sigma^2. \]Total loss
\[ S_n=X_1+\cdots+X_n \]has
\[ E[S_n]=n\mu \]and
\[ \operatorname{Var}(S_n)=n\sigma^2. \]The average loss
\[ \bar X_n=\frac{S_n}{n} \]has variance
\[ \boxed{ \operatorname{Var}(\bar X_n)=\frac{\sigma^2}{n} }. \]Pooling Does Not Eliminate All Risk
Pooling is less effective when losses are strongly correlated.
Examples include losses caused by:
- regional disasters;
- market crashes;
- system-wide cyber events;
- common legal changes;
- shared infrastructure failure.
Dependence is therefore central to risk modeling.
Diversification
Diversification combines different exposures so that poor outcomes in one position may be offset by better outcomes elsewhere.
Its effectiveness depends on dependence among returns or losses.
Combining two risky assets does not automatically reduce risk if they move together perfectly.
Portfolio Return
Suppose a portfolio places weight \(w_i\) in asset \(i\), with
\[ \sum_i w_i=1. \]If asset returns are \(R_i\), portfolio return is
\[ \boxed{ R_P=\sum_i w_iR_i }. \]Therefore,
\[ E[R_P] = \sum_i w_iE[R_i]. \]Two-Asset Portfolio Variance
For two assets with weights \(w_1,w_2\),
\[ \boxed{ \operatorname{Var}(R_P) = w_1^2\sigma_1^2 + w_2^2\sigma_2^2 + 2w_1w_2\operatorname{Cov}(R_1,R_2) }. \]The covariance term shows why dependence matters.
Correlation
Correlation standardizes covariance:
\[ \boxed{ \rho_{12} = \frac{\operatorname{Cov}(R_1,R_2)} {\sigma_1\sigma_2} }. \]Lower correlation can increase diversification benefit, all else equal.
Independent Does Not Mean Risk-Free
Two independent investments can each remain individually risky.
Independence helps reduce certain forms of aggregate variability, but it does not make every outcome predictable.
Risk Measures
Different risk measures answer different questions.
Examples include:
- variance;
- standard deviation;
- probability of loss;
- expected loss;
- quantiles;
- tail expectation;
- probability of ruin.
No single risk measure captures every aspect of uncertainty.
Probability of Loss
If \(X\) is net financial gain, then
\[ \boxed{P(X<0)} \]is the probability of a loss.
This can differ substantially between choices with the same expected value.
Quantiles
A \(q\)-quantile is a threshold below which a specified fraction \(q\) of outcomes lie.
Quantiles help summarize the lower or upper tail of a distribution.
In loss modeling, an upper loss quantile answers a question such as:
what loss level is exceeded only a small fraction of the time?
Value at Risk as a Quantile Concept
One common financial risk concept, Value at Risk, is based on a loss quantile at a chosen confidence level.
In a simplified notation, if \(L\) is loss, then
\[ \boxed{ \operatorname{VaR}_{\alpha}(L) = \text{an }\alpha\text{-quantile of }L }. \]Precise conventions matter, especially for discrete distributions.
Tail Risk
A quantile does not fully describe what happens beyond the threshold.
Two loss distributions may have the same high quantile but very different extreme losses beyond it.
Tail-sensitive measures examine the severity of outcomes in the worst part of the distribution.
Expected Shortfall Idea
Expected shortfall is designed to summarize losses in the tail beyond a specified confidence threshold.
In continuous settings, one intuitive interpretation is:
\[ \boxed{\text{average loss among sufficiently severe tail outcomes}}. \]Exact definitions require care for discrete distributions and boundary cases.
Utility and Risk Aversion
Expected monetary value does not always capture how people value risky outcomes.
A utility function
\[ u(x) \]represents preference over wealth or payoff.
A risk-averse decision maker typically has a concave utility function.
Expected Utility
Under expected-utility reasoning, a risky alternative is evaluated using
\[ \boxed{ E[u(X)] = \sum_i p_i u(x_i) }. \]Maximizing expected utility can produce a different decision from maximizing expected monetary payoff.
Why Insurance Can Be Valuable
Insurance can exchange a small certain cost for protection against a large uncertain loss.
A risk-averse person may prefer this trade even when the premium exceeds expected claim cost.
The difference reflects both insurer expenses and the value placed on risk reduction.
Risk-Neutral Decision Rule
A simplified risk-neutral rule chooses the alternative with the highest expected monetary payoff.
If alternatives \(A\) and \(B\) satisfy
\[ E[X_A]>E[X_B], \]a risk-neutral expected-value decision maker prefers A.
Risk Aversion
A risk-averse decision maker may prefer a lower-mean alternative if it sufficiently reduces uncertainty or severe downside.
Therefore,
\[ \boxed{\text{higher expected return does not automatically mean preferred}} \]when risk matters.
Risk Premium
In utility-based decision theory, a risk premium is the amount of expected value a decision maker would give up to avoid a risk.
This idea formalizes the value of certainty or risk reduction.
Ruin
In insurance and finance, ruin describes a state in which available capital falls below a required threshold, often zero.
If surplus evolves as
\[ U_t = U_0+\text{premium income}-\text{claims}, \]ruin occurs if
\[ U_t<0 \]at some relevant time.
Ruin Probability
The probability
\[ \boxed{ P(\text{ruin occurs}) } \]depends on:
- initial capital;
- premium income;
- claim frequency;
- claim severity;
- time horizon;
- dependence;
- investment returns.
Rare Events and Severe Losses
A low-probability event can still dominate risk if the potential loss is enormous.
Expected loss multiplies probability and severity:
\[ \boxed{\text{expected contribution}=p\times\text{loss amount}}. \]But expected value alone may still understate concern about catastrophic tail outcomes.
Scenario Analysis
A practical risk analysis often compares several scenarios.
| Scenario | Probability | Outcome |
|---|---|---|
| Favorable | 0.25 | \(\$500\) |
| Normal | 0.55 | \(\$100\) |
| Adverse | 0.20 | \(-\$600\) |
Expected payoff is
\[ 0.25(500)+0.55(100)+0.20(-600). \]Therefore,
\[ 125+55-120 = \boxed{\$60}. \]Sensitivity Analysis
Risk estimates depend on uncertain parameters.
Sensitivity analysis asks how conclusions change when assumptions change.
Examples include changing:
- claim probability;
- loss severity;
- return probabilities;
- correlations;
- deductibles;
- capital levels.
Stress Testing
Stress testing evaluates a system under unusually severe but plausible or deliberately extreme conditions.
It differs from ordinary expected-value analysis because it focuses on adverse scenarios rather than average behavior.
Simulation in Risk Analysis
Monte Carlo simulation can approximate the distribution of complex financial or insurance outcomes.
A simulation may generate:
- random returns;
- claim counts;
- claim severities;
- portfolio values;
- capital paths;
- ruin events.
Monte Carlo Workflow
- Specify the probability model.
- Generate one random scenario.
- Calculate the resulting financial outcome.
- Repeat many times.
- Summarize the simulated distribution.
Useful summaries include means, standard deviations, quantiles, tail losses, and event probabilities.
Model Risk
Model risk is the risk that the chosen mathematical model does not adequately represent reality.
Sources include:
- incorrect assumptions;
- poor parameter estimates;
- structural breaks;
- ignored dependence;
- rare events absent from historical data.
Parameter Uncertainty
A probability such as
\[ p=0.02 \]may itself be estimated from data.
Therefore, there can be uncertainty both:
- inside the stochastic model;
- about the model parameters themselves.
Common Error: Treating Expected Value as a Certain Outcome
If
\[ E[X]=\$100, \]that does not mean the realized gain will be \(\$100\).
Expected value summarizes the distribution.
Common Error: Ignoring the Sign Convention
Gains and losses must be represented consistently.
If gains are positive and losses are negative, a \(\$500\) loss should enter an expected-payoff calculation as
\[ \boxed{-500}. \]Common Error: Comparing Expected Returns Without Risk
Two investments can have identical expected returns but very different variances and downside probabilities.
Expected return alone is not a complete description of a risky choice.
Common Error: Assuming Diversification Always Works
Diversification depends on dependence.
Highly correlated assets or losses can move together and provide little diversification benefit.
Common Error: Ignoring Contract Terms
Insurance loss \(L\) is not always equal to insurer payment.
Deductibles, limits, coinsurance, exclusions, and other provisions alter payment.
The random variable must match the quantity actually being modeled.
Common Error: Treating a Quantile as a Maximum Loss
A high loss quantile is not necessarily the worst possible loss.
Losses beyond the quantile can still occur.
Tail analysis is needed to understand those more extreme outcomes.
Common Error: Assuming Historical Probabilities Are Permanent
Financial and insurance environments can change.
Past frequencies may become poor estimates after changes in:
- technology;
- law;
- behavior;
- climate;
- markets;
- policy terms.
A Reliable Procedure for Expected-Value Problems
- List the possible outcomes.
- Assign probabilities.
- Verify that probabilities sum to \(1\).
- Choose a consistent gain/loss sign convention.
- Multiply each outcome by its probability.
- Add the probability-weighted values.
- Interpret the result as an expectation, not a guarantee.
A Reliable Procedure for Insurance-Loss Problems
- Define the loss random variable.
- Separate claim frequency from severity when useful.
- Identify deductibles and limits.
- Transform loss into insurer payment if required.
- Calculate expected payment.
- Measure variability or tail risk when relevant.
- Check independence and dependence assumptions.
- Compare expected claims with capital and premium structure.
A Reliable Procedure for Portfolio-Risk Problems
- List the assets and portfolio weights.
- Verify that weights sum to \(1\) unless leverage or cash is modeled separately.
- Calculate expected asset returns.
- Calculate portfolio expected return.
- Identify variances and covariances.
- Calculate portfolio variance.
- Examine downside probabilities and tail measures if relevant.
- Test sensitivity to correlations and scenario assumptions.
Example: Full Insurance Workflow
Suppose a policy has the following annual loss distribution:
| Loss | Probability |
|---|---|
| \(\$0\) | 0.85 |
| \(\$2{,}000\) | 0.10 |
| \(\$10{,}000\) | 0.05 |
Expected loss:
\[ E[L] = 0.85(0)+0.10(2000)+0.05(10000). \]Therefore,
\[ E[L] = 200+500 = \boxed{\$700}. \]If a \(\$1{,}000\) deductible applies, insurer payments become:
\[ 0,\qquad1000,\qquad9000. \]Expected insurer payment is
\[ 0.85(0)+0.10(1000)+0.05(9000) = 100+450 = \boxed{\$550}. \]Example: Full Portfolio Workflow
Suppose a portfolio invests
\[ 60\% \]in Asset A and
\[ 40\% \]in Asset B.
Expected returns are
\[ E[R_A]=10\% \]and
\[ E[R_B]=5\%. \]Portfolio expected return is
\[ E[R_P] = 0.6(0.10)+0.4(0.05). \]Therefore,
\[ E[R_P] = 0.06+0.02 = \boxed{0.08=8\%}. \]Example: Full Risk-Comparison Workflow
Investment A always returns
\[ 4\%. \]Investment B returns
\[ 12\% \]with probability \(0.5\) and
\[ -4\% \]with probability \(0.5\).
Expected return of B is
\[ 0.5(0.12)+0.5(-0.04) = 0.06-0.02 = \boxed{0.04}. \]Both investments have expected return \(4\%\).
Investment A has zero return variance in this simplified model, while B has positive variance.
Therefore, expected return alone does not distinguish them.
Practice 1: Expected Payoff
A gamble pays \(\$300\) with probability \(0.4\) and loses \(\$100\) with probability \(0.6\).
Find the expected payoff.
Practice 2: Simple Return
An asset rises from \(\$80\) to \(\$92\).
Find the simple return.
Practice 3: Expected Loss
A loss of \(\$5{,}000\) occurs with probability \(0.03\); otherwise the loss is zero.
Find expected loss.
Practice 4: Deductible
A covered loss is \(\$7{,}500\) and the deductible is \(\$1{,}500\).
Find the insurer payment when there is no additional policy limit.
Practice 5: Claim Frequency and Severity
A portfolio expects \(25\) claims with mean severity \(\$1{,}200\).
Under the simple compound expectation rule, find expected aggregate loss.
Practice 6: Portfolio Expected Return
A portfolio places \(70\%\) in an asset with expected return \(9\%\) and \(30\%\) in an asset with expected return \(3\%\).
Find the portfolio expected return.
Practice 7: Probability of Loss
A random payoff is negative in \(18\%\) of scenarios.
What is \(P(X<0)\)?
Practice 8: Pooling Variance
Independent losses have variance
\[ \sigma^2=100. \]If \(25\) identical independent risks are averaged, find the variance of the average.
Practice 9: Ruin Condition
If surplus is
\[ U_t=-\$200, \]has ruin occurred under the simple zero-threshold definition?
Practice 10: Diversification
Does combining two risky assets automatically eliminate risk?
Practice Answers
Practice 1
\[ E[X] = 0.4(300)+0.6(-100) = 120-60 = \boxed{\$60}. \]Practice 2
\[ R = \frac{92-80}{80} = \frac{12}{80} = \boxed{0.15=15\%}. \]Practice 3
\[ E[L] = 0.03(5000)+0.97(0) = \boxed{\$150}. \]Practice 4
\[ 7500-1500 = \boxed{\$6{,}000}. \]Practice 5
\[ E[S] = 25(1200) = \boxed{\$30{,}000}. \]Practice 6
\[ E[R_P] = 0.7(0.09)+0.3(0.03) = 0.063+0.009 = \boxed{0.072=7.2\%}. \]Practice 7
\[ \boxed{P(X<0)=0.18}. \]Practice 8
\[ \operatorname{Var}(\bar X) = \frac{100}{25} = \boxed{4}. \]Practice 9
Yes.
Since
\[ U_t<0, \]ruin has occurred under the stated definition.
Practice 10
No.
Diversification can reduce some risk, especially when exposures are not perfectly positively dependent, but it does not automatically eliminate all risk.
What to Remember
Finance, insurance, and risk models use probability distributions to compare uncertain gains and losses.
- expected value is a probability-weighted average;
- returns measure change relative to starting value;
- variance and standard deviation measure variability;
- insurance separates claim frequency from claim severity;
- deductibles and limits transform losses into payments;
- risk pooling can make average losses more predictable;
- diversification depends on covariance and correlation;
- quantiles summarize distribution thresholds;
- tail measures examine severe outcomes beyond ordinary averages;
- expected utility can model risk aversion;
- ruin probability focuses on capital falling below a critical threshold;
- simulation helps analyze complex loss and return distributions.
The central habit is: describe the entire distribution of possible outcomes, not just the average, and keep probability, severity, dependence, and decision preferences separate.
Where We Go Next
Finance, insurance, and risk models show how probability supports decisions involving uncertain gains and losses. The next applied problem family uses the same conditional-probability ideas in settings where test results and population prevalence must be translated into evidence about disease or public-health status.
The next lesson, Medical Testing and Public-Health Inference, develops sensitivity, specificity, prevalence, predictive values, Bayes' theorem, screening, sampling, and population-level interpretation.