Quantitative and Valuation Foundations

Introduction and Context

Investment banking decisions depend on translating uncertain future cash flows and market outcomes into measurable values. Probability distributions help describe uncertainty; correlation and volatility describe dependence and dispersion; time-series methods organize information through time; and valuation techniques convert future cash flows into present economic values. These tools support pricing, hedging, capital allocation, transaction analysis, market risk measurement and financial reporting.

Quantitative models do not eliminate uncertainty. They impose a structured representation on incomplete and noisy information. A result is therefore meaningful only when its assumptions, data quality, model specification and estimation error are understood. A precise-looking valuation can still be economically unreliable if the discount rate, cash-flow forecast or market input is inappropriate.

Learning Objectives

By the end of this chapter, the reader should be able to:

1. Explain probability distributions and select appropriate measures of location, dispersion and tail risk.
2. Calculate covariance, correlation, portfolio volatility and interpret diversification.
3. Distinguish returns, volatility, stationarity, autocorrelation and basic time-series models.
4. Apply discounting and present-value methods to bonds, projects and transactions.
5. Explain mark-to-market valuation, fair-value hierarchy and valuation adjustments.
6. Calculate and interpret duration, convexity and price sensitivities.
7. Design scenario and sensitivity analyses while recognizing model and data limitations.
8. Communicate quantitative results responsibly in investment-banking practice.

1. Probability and Distributions in Risk Analysis

1.1 Random variables and probability

A random variable assigns numerical outcomes to uncertain events. A discrete variable has countable outcomes, whereas a continuous variable can take values over an interval. For a discrete variable X, its probability mass function satisfies:

The expected value is the probability-weighted average:

For a continuous variable with density f(x):

The variance measures squared dispersion around the mean μ:

Standard deviation is σ = √Var(X). If X is measured in currency, variance is in squared currency units, while standard deviation is in currency units. For returns, standard deviation is expressed as a proportion or percentage.

The cumulative distribution function is F(x) = P(X ≤ x). It is useful for estimating probabilities such as the chance that a loss exceeds a specified threshold.

1.2 Common distributions

The normal distribution is symmetric and described by mean μ and standard deviation σ. It is often useful for aggregated market changes, but it understates skewness and extreme events when used uncritically. A standardized normal variable is:

A Bernoulli variable represents one trial with success probability p. A binomial variable counts successes in n independent Bernoulli trials. A lognormal variable is always positive and is frequently used for asset prices when continuously compounded returns are modeled as normal.

The Student’s t-distribution has heavier tails than the normal distribution and can be useful when samples are small or returns exhibit excess kurtosis. Empirical return distributions may also be skewed, exhibit volatility clustering and contain jumps; no single standard distribution should be assumed without diagnostic testing.

1.3 Quantiles and downside measures

A quantile qα is a value such that a specified proportion α of observations lies below it. For a loss variable L, Value at Risk at confidence level α is commonly defined as the α-quantile of loss:

VaR answers a percentile question but does not describe the average loss beyond that percentile. Expected Shortfall (ES), also called conditional VaR, does so:

Definitions must specify whether L represents a positive loss or a negative return. Confusing profit-and-loss conventions can reverse conclusions.

2. Dependence, Correlation and Portfolio Volatility

2.1 Covariance and correlation

For random variables X and Y, covariance is:

Sample covariance for n observations is:

Correlation standardizes covariance:

It lies between −1 and +1. Correlation measures linear dependence, not causation or complete dependence. Two assets may have low correlation in normal markets but become strongly related during crises. Correlation is also sensitive to the observation frequency, sample period and non-synchronous prices.

2.2 Portfolio variance

For a two-asset portfolio with weights w₁ and w₂, returns R₁ and R₂, and standard deviations σ₁ and σ₂:

Weights may be positive or negative, and typically sum to one for a fully invested portfolio. The covariance-matrix form for n assets is:

where w is the n×1 weight vector and Σ is the n×n covariance matrix. The matrix must be symmetric and positive semidefinite for a valid variance model.

Worked example. Suppose a portfolio invests 60% in Asset A and 40% in Asset B. Their annual volatilities are 20% and 10%, and correlation is 0.25. Then:

σₚ² = (0.6²)(0.2²) + (0.4²)(0.1²) + 2(0.6)(0.4)(0.25)(0.2)(0.1)

= 0.0144 + 0.0016 + 0.0024 = 0.0184.

Therefore, σₚ = √0.0184 ≈ 0.1356, or approximately 13.56% annually. This is below the weighted average of individual volatilities because diversification reduces risk when correlation is below one.

2.3 Diversification and concentration

Diversification is not equivalent to holding many securities. Effective diversification depends on exposures, common risk factors, liquidity, leverage and dependence in stressed conditions. A portfolio may contain many issuers but remain concentrated in one sector, currency, country or funding source. Risk reports should therefore supplement asset counts with exposure-based concentration measures and stress correlations.

3. Volatility and Time-Series Concepts

3.1 Returns and compounding

For price Pₜ, the simple return is:

The continuously compounded, or log, return is:

Log returns add over time, whereas simple returns compound multiplicatively. For periods with prices P₀ and P_T, cumulative simple return is P_T ÷ P₀ − 1. Annualization must match the data frequency. Under an assumption of independent and identically distributed returns, annualized volatility from k observations per year is:

This square-root-of-time rule is an approximation and fails when returns are autocorrelated, volatility is time-varying or the horizon contains jumps.

3.2 Stationarity and autocorrelation

A weakly stationary series has a constant mean and variance and an autocovariance that depends on lag rather than calendar time. Asset prices are generally non-stationary, while returns are often treated as approximately stationary over limited periods. This distinction matters because regressions involving unrelated trending prices can produce spurious relationships.

Autocorrelation at lag k measures dependence between a series and its k-period lagged value. Positive autocorrelation can arise from illiquidity or smoothing; negative autocorrelation can arise from bid–ask bounce or short-term reversal. Time-series diagnostics should include plots, rolling statistics, autocorrelation functions and tests appropriate to the data.

An autoregressive model of order one, AR(1), is:

where c is a constant, φ is the autoregressive coefficient and εₜ is an innovation with mean zero. A sufficient condition for covariance stationarity in the basic AR(1) model is |φ| < 1. Its unconditional mean is c ÷ (1 − φ), provided the process is stationary.

3.3 Conditional volatility

Financial returns often show volatility clustering: large changes tend to follow large changes. A simple GARCH(1,1) variance specification is:

where ω > 0, α ≥ 0 and β ≥ 0. A commonly required covariance-stationarity condition is α + β < 1. The model assumes that current variance depends on a long-run component, recent squared shocks and recent variance. It does not automatically capture jumps, leverage effects or structural breaks.

4. Discounting and Present Value

4.1 Time value of money

A unit of currency today generally differs from a unit received later because of opportunity cost, inflation, risk and liquidity. At an effective per-period discount rate r, the present value of a cash flow CFₜ received at time t is:

For multiple cash flows:

Using a term structure, rₜ must correspond to the maturity of each cash flow. A zero-coupon discount factor DFₜ provides a more direct expression:

The rate convention must match the cash-flow frequency. A nominal annual rate compounded m times per year has per-period rate r_nominal ÷ m, while its effective annual rate is:

4.2 Net present value and internal rate of return

For an investment with initial outlay CF₀ and future cash flows:

A positive NPV indicates value creation relative to the specified discount rate, assuming cash flows and rate are consistently defined. The internal rate of return is the rate r that sets NPV equal to zero. IRR can be misleading for non-conventional cash flows, multiple sign changes or mutually exclusive projects; NPV is generally the stronger decision criterion when the discount rate is defensible.

4.3 Bond valuation

For a fixed-rate bond with coupon C per period, face value F, yield y per period and n periods:

If coupons are annual, C = F × coupon rate. For semiannual coupons, the coupon and yield must be expressed per half-year and n must count half-year periods. A bond trades at par when its yield equals its coupon rate under consistent conventions.

Worked example. A three-year annual bond has face value 1,000, coupon rate 5% and yield 6%. Its price is:

P = 50 ÷ 1.06 + 50 ÷ 1.06² + 1,050 ÷ 1.06³

≈ 47.17 + 44.50 + 881.67 = 973.34.

The price is below par because the market yield exceeds the coupon rate.

5. Mark-to-Market Valuation and Fair Value

Mark-to-market (MTM) valuation measures an instrument using current market information rather than historical cost. When an active-market quoted price exists for an identical instrument, it is generally the strongest evidence. Where no direct quote exists, valuation may use comparable instruments, observable curves, credit spreads, option-implied inputs or models.

A practical valuation hierarchy is:

For a derivative, fair value is often the discounted expected value of future payoffs under a pricing measure, adjusted for collateral, funding, credit and other relevant effects according to the applicable accounting and risk framework. The exact treatment depends on instrument type and governing standards.

MTM gains and losses may arise from market movements, passage of time, accrual, changes in credit spreads, foreign exchange and model-input updates. A valuation control process should include independent price verification, valuation reserves where appropriate, observable-input testing, limit monitoring, documentation and escalation of stale or disputed prices.

6. Duration, Convexity and Sensitivities

6.1 Macaulay and modified duration

For a bond with cash flow CFₜ and yield y per period, Macaulay duration is the present-value-weighted average time to receipt:

where P is the bond price and t is measured in coupon periods. Modified duration for a yield quoted per period is:

For a small yield change Δy, the first-order price approximation is:

The units of D_mod are periods per unit yield. If yield changes by 0.01, that is a one-percentage-point change, not one basis point. For a one-basis-point change, Δy = 0.0001.

6.2 Convexity

Duration assumes a linear price–yield relationship. Convexity captures curvature. A standard discrete convexity measure is:

The second-order approximation is:

This formula applies under consistent periodic conventions and small yield changes. Positive convexity generally improves the approximation and means that price rises from a yield fall exceed the price loss from an equal yield rise, all else equal. Callable bonds may exhibit negative effective convexity because issuer exercise behavior changes the cash-flow profile.

6.3 Key-rate and spread sensitivities

A single duration number can conceal exposure to different maturities. Key-rate duration measures price sensitivity to a localized movement at a specific point on the yield curve while other key rates are held constant. Spread duration measures sensitivity to a credit-spread change. DV01, or dollar value of a one-basis-point move, can be approximated by:

This approximation assumes a parallel yield movement and a price convention consistent with the calculation. For a portfolio, sensitivities should be aggregated by sign, maturity, currency, curve and risk factor rather than simply summed in absolute value.

7. Sensitivity and Scenario Analysis

Sensitivity analysis changes one input or a small group of inputs while holding other assumptions constant. Examples include changing a discount rate, terminal growth rate, default probability, exit multiple or foreign-exchange rate. It identifies value drivers but may understate risk when variables move together.

Scenario analysis changes a coherent set of assumptions. A banking scenario might combine higher interest rates, wider credit spreads, lower volumes, weaker collateral values and increased funding costs. Scenarios should include a base case, adverse case and severe but plausible case. Reverse stress testing starts with an unacceptable outcome and asks which combination of conditions could produce it.

A useful scenario workflow is:

1. Define the decision, portfolio and risk horizon.
2. Identify material risk factors and dependencies.
3. Select historical, hypothetical or hybrid shocks.
4. Map shocks to cash flows, prices, collateral and liquidity.
5. Revalue positions, including nonlinear effects where material.
6. Compare results with limits, capital and liquidity resources.
7. Document assumptions, uncertainty, management actions and governance.

Sensitivity results are not forecasts. They are conditional statements: “If these assumptions hold, the estimated value changes by this amount.”

Illustrative Case: Valuing and Stressing a Corporate Bond Position

Background

An investment bank holds a three-year corporate bond with face value 10 million, annual coupon rate 5%, and market yield of 6%. The position is used in a client-financing book. The risk team must estimate its value, rate sensitivity and a stress loss.

Analysis

Using the bond formula, the price per 1,000 face value is approximately 973.34. The position’s market value is therefore approximately:

10,000,000 × 973.34 ÷ 1,000 = 9,733,400.

The bond’s Macaulay duration is approximately 2.82 years, giving modified duration approximately:

D_mod = 2.82 ÷ 1.06 ≈ 2.66.

For a 25-basis-point parallel yield increase, Δy = 0.0025. The duration-only estimated percentage price change is:

ΔP ÷ P ≈ −2.66 × 0.0025 = −0.00665,

or −0.665%. The estimated loss is approximately 64,700 before convexity, spread effects, transaction costs and hedging.

Suppose the stress also widens the credit spread by 100 basis points. If spread duration is estimated at 2.4, the additional approximate loss is 2.4% of market value, or about 233,600. The combined estimate is not necessarily the sum of independent risks in a nonlinear model; it should be recalculated using a full revaluation where possible.

Approach and outcome

The risk team performs a full repricing using the stressed risk-free curve and credit spread, compares the result with the duration approximation, and reports both. It verifies that the bond is not callable, confirms the coupon frequency, tests the liquidity of the observed yield, and applies an independent price verification process. The result is presented with a range rather than false precision because spread duration, liquidity and recovery assumptions are estimated.

Learning points

The case illustrates that market value is distinct from accounting cost, duration is a local approximation, credit and interest-rate risks can interact, and a stress result is conditional on model inputs. Governance requires transparent assumptions and an independent challenge function.

Limitations, Governance and Professional Practice

Quantitative estimates are limited by short samples, survivorship bias, stale prices, bid–ask spreads, regime changes, missing data, look-ahead bias and unstable correlations. Historical simulation assumes that the past contains relevant future shocks; parametric models impose distributional assumptions; Monte Carlo methods can produce highly precise results from a misspecified model.

Model risk management should include documented purpose, conceptual soundness review, data lineage, validation, benchmarking, sensitivity testing, performance monitoring, approval authority and clear restrictions on use. Estimates should distinguish market-observable inputs from management assumptions.

Ethically, analysts must not select scenarios solely to support a transaction, suppress adverse valuation evidence, misuse confidential information or present model outputs as objective facts. Independent valuation, segregation of duties and audit trails protect clients, counterparties and the institution. Regulatory requirements vary by jurisdiction, but governance commonly emphasizes prudent valuation, fair-value disclosure, capital adequacy, stress testing and record retention.

Technology can improve data collection, automated repricing and anomaly detection. Machine-learning models may identify nonlinear patterns, but they introduce risks involving explainability, drift, data leakage, bias and overfitting. AI-generated analysis requires human validation, controlled data access and reproducibility. Automation should strengthen—not replace—professional judgment.

Best-Practice Checklist

• Match return, yield and discount-rate conventions to the cash-flow frequency.
• Use market-observable inputs where available and document adjustments.
• Validate units, signs, compounding and horizon conversions.
• Report both central estimates and uncertainty ranges.
• Test non-normality, volatility clustering and changing correlations.
• Use full revaluation for material nonlinear positions.
• Distinguish sensitivity analysis from probabilistic forecasting.
• Challenge data quality, model assumptions and incentives.
• Escalate stale prices, valuation disputes and breaches promptly.

Chapter Summary

Probability distributions provide a language for uncertain outcomes, while variance, standard deviation and quantiles summarize risk. Covariance and correlation explain dependence, and the covariance-matrix portfolio formula shows why diversification depends on relationships rather than security count. Returns must be distinguished from prices, and volatility estimates require attention to compounding, stationarity, autocorrelation and clustering.

Discounting converts future cash flows into present value and requires consistent timing, rate and compounding conventions. Mark-to-market valuation uses current market evidence, with increasing model uncertainty when observable inputs are unavailable. Duration estimates first-order bond-price sensitivity, convexity improves the approximation for larger movements, and key-rate, spread and DV01 measures reveal more specific exposures. Scenario analysis evaluates conditional outcomes and should include coherent, severe-but-plausible combinations of shocks.

The central professional lesson is that quantitative outputs are estimates, not facts. Sound investment-banking practice combines mathematically correct calculations with data governance, independent validation, transparent assumptions, ethical judgment and effective communication of uncertainty.

Key Terms

Probability distribution: A rule describing the likelihood of possible outcomes.

Expected value: Probability-weighted average outcome.

Variance: Expected squared deviation from the mean.

Standard deviation: Square root of variance and a measure of dispersion.

Quantile: Value below which a specified proportion of observations falls.

Value at Risk: Loss quantile at a stated confidence level and horizon.

Expected Shortfall: Average loss conditional on exceeding a VaR threshold.

Covariance: Measure of joint variation between two variables.

Correlation: Standardized measure of linear dependence.

Volatility: Standard deviation of returns, usually over a specified horizon.

Stationarity: Stability of key statistical properties through time.

Autocorrelation: Dependence between observations separated by a specified lag.

Discount factor: Present value of one unit received at a future date.

Net present value: Discounted value of future cash flows including initial investment.

Mark-to-market: Valuation using current market information.

Duration: First-order sensitivity of a bond’s price to yield.

Convexity: Curvature of the bond price–yield relationship.

DV01: Approximate price or value change for a one-basis-point yield movement.

Sensitivity analysis: Conditional analysis changing selected inputs.

Scenario analysis: Analysis of a coherent set of changed assumptions.

Review Questions

Short Answer Questions

1. Distinguish variance from standard deviation and explain their units.
2. Why does correlation affect portfolio volatility?
3. State two conditions required for a basic AR(1) process to be stationary.
4. Explain why a bond trades below par when its yield exceeds its coupon rate.
5. What information does convexity add to duration?

Long Answer Questions

1. Explain the strengths and limitations of normal, lognormal and Student’s t-distributions in market-risk analysis.
2. Derive and interpret the two-asset portfolio variance formula, including the role of diversification.
3. Discuss discounting using a term structure and explain why rate conventions must match cash-flow frequency.
4. Compare mark-to-market valuation based on quoted prices, observable inputs and model-based inputs.
5. Explain how model risk, data limitations and governance affect the reliability of quantitative valuation.

Analytical / Application Questions

1. A portfolio has weights 0.7 and 0.3, volatilities 18% and 12%, and correlation −0.10. Calculate its volatility and interpret the diversification effect.
2. A bond has modified duration 4.5 and convexity 30. Estimate the percentage price change for a 75-basis-point yield increase using the duration–convexity approximation.
3. Design a stress scenario for a leveraged acquisition financing portfolio. Identify at least four linked shocks, explain their transmission channels and state which positions require full revaluation.

Multiple Choice Questions with Answers

Question 1

Which measure represents the average loss conditional on losses exceeding a specified VaR threshold?

A. Variance

B. Expected Shortfall

C. Correlation

D. Duration

Correct Answer: B. Expected Shortfall

Question 2

A portfolio has two assets with correlation below +1. Holding other inputs constant, what is the usual effect?

A. Portfolio volatility is reduced through diversification

B. Both asset volatilities become zero

C. Portfolio variance must equal the larger asset variance

D. Expected return becomes certain

Correct Answer: A. Portfolio volatility is reduced through diversification

Question 3

For an AR(1) process Xₜ = c + φXₜ₋₁ + εₜ, which condition supports covariance stationarity in the basic model?

A. φ = 2

B. |φ| < 1

C. φ > 1

D. εₜ must be positive

Correct Answer: B. |φ| < 1

Question 4

A bond’s coupon rate is 5% and its yield is 6%. Under consistent annual conventions, the bond will generally trade:

A. Above par

B. At exactly twice par

C. Below par

D. Only at zero

Correct Answer: C. Below par

Question 5

A bond has modified duration 4. If its yield rises by 0.01, the duration-only estimated percentage price change is:

A. +4%

B. −4%

C. −0.04%

D. +0.04%

Correct Answer: B. −4%

Question 6

What does convexity primarily capture in fixed-income analysis?

A. The issuer’s accounting profit

B. Curvature in the price–yield relationship

C. The number of bondholders

D. The bond’s coupon frequency only

Correct Answer: B. Curvature in the price–yield relationship

Question 7

Which expression is the present value of a single cash flow CFₜ discounted at rate r per period?

A. CFₜ × (1 + r)ᵗ

B. CFₜ ÷ (1 + r)ᵗ

C. CFₜ + r + t

D. CFₜ ÷ rᵗ only

Correct Answer: B. CFₜ ÷ (1 + r)ᵗ

Question 8

What is a principal limitation of correlation?

A. It can measure only units of currency

B. It proves causation

C. It measures linear dependence and may change in stressed markets

D. It is always greater than one

Correct Answer: C. It measures linear dependence and may change in stressed markets

Question 9

Sensitivity analysis differs from scenario analysis because sensitivity analysis typically:

A. Changes one input or a small group while holding others constant

B. Always uses historical crises

C. Requires a probability distribution for every outcome

D. Cannot be used for valuation

Correct Answer: A. Changes one input or a small group while holding others constant

Question 10

A DV01 approximation is based primarily on which inputs?

A. Coupon colour and issuer name

B. Price, modified duration and 0.0001

C. Face value only

D. Correlation and skewness only

Correct Answer: B. Price, modified duration and 0.0001