Quantitative Methods for Investment Decisions

Introduction

Investment management converts uncertain future cash flows into present decisions. Quantitative methods provide the language for doing so: time value of money compares cash flows occurring at different dates; probability represents uncertainty; statistics summarizes evidence; correlation and regression explain relationships; scenario analysis and simulation explore possible futures; and downside-risk measures focus attention on harmful outcomes.

These methods are powerful but not self-validating. A precisely calculated result can still be misleading when the data are biased, the model is misspecified, the assumptions are unrealistic, or the decision-maker mistakes historical regularities for laws of nature. Professional investment analysis therefore combines mathematical discipline with economic judgment, governance and sensitivity testing.

Learning Objectives

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

1. Apply present-value, future-value, annuity and internal-rate-of-return methods.
2. Use probability distributions, expected value and variance to describe investment uncertainty.
3. Estimate sample statistics and construct confidence intervals.
4. Interpret covariance, correlation and regression in portfolio analysis.
5. Perform scenario analysis and explain the logic of Monte Carlo simulation.
6. Calculate and interpret downside-risk measures, including semivariance, value at risk and conditional value at risk.
7. Evaluate model risk, data limitations, ethical issues and appropriate governance controls.

1. Time Value of Money

A monetary unit available today can be invested and may earn a return. Consequently, a certain amount today is generally worth more than the same nominal amount received later. The time value of money (TVM) is the foundation of bond valuation, equity valuation, capital budgeting and performance measurement.

1.1 Future and present value

With annual compounding, the future value of a present amount is:

where PV is present value, FV is future value, r is the effective return per period, and n is the number of periods. The rate and number of periods must refer to the same period. The inverse relationship is:

For cash flows CF_t received at dates t = 1, 2, ..., n, the value at date zero is:

This formula assumes a constant discount rate and cash flows known or estimated at the relevant dates. If rates vary by maturity, each cash flow should be discounted using its appropriate spot rate s_t:

1.2 Effective and nominal rates

If an annual nominal rate j is compounded m times per year, the effective annual rate is:

For example, a 12% nominal rate compounded monthly has EAR = (1 + 0.12 ÷ 12)¹² − 1 ≈ 12.68%. Discounting a monthly cash flow requires a monthly rate, not an unadjusted annual rate. Unit consistency is essential.

1.3 Annuities and perpetuities

An ordinary annuity pays the same amount C at the end of each period. Its present value is:

Its future value at the end of period n is:

An annuity due pays at the beginning of each period. Its value is the ordinary-annuity value multiplied by (1 + r). A level perpetuity has no terminal payment date and is valued, when r > 0, as:

A growing perpetuity with payment next period C₁ and constant growth g is:

The growing-perpetuity formula requires r > g, a constant growth rate and a discount rate appropriate to the risk of the cash flows. It is not suitable for a business whose growth is expected to exceed the discount rate indefinitely.

Worked example: project valuation

An investment requires £100,000 today and is expected to pay £45,000 at the end of each of the next three years. At a discount rate of 10%:

The net present value is:

Subject to the assumptions, the project adds value at the 10% required return. NPV is generally preferred to percentage returns for mutually exclusive projects because it measures value added in currency and is additive across projects.

1.4 Internal rate of return and limitations

The internal rate of return (IRR) is the discount rate that makes NPV zero:

IRR can be useful for communication, but it may produce multiple rates when cash flows change sign more than once, and it can rank mutually exclusive projects incorrectly. It also implicitly assumes reinvestment at the IRR, which may be unrealistic. Modified IRR addresses reinvestment assumptions, while NPV remains the stronger primary decision criterion when the required return is known.

2. Probability and Investment Uncertainty

A probability model assigns probabilities to possible states of the world. A discrete random variable takes countable outcomes; a continuous random variable can take values over an interval. Probabilities must be non-negative and sum, or integrate, to one.

For possible returns R_i with probabilities p_i, expected return is:

Expected return is a probability-weighted average, not necessarily the return that will occur. Variance measures dispersion:

Standard deviation is σ(R) = √σ²(R). If returns are expressed as decimals, variance has squared units and standard deviation has the same units as return. Standard deviation treats positive and negative deviations symmetrically; investors often need additional downside measures.

Example: expected return and risk

Suppose an asset has the following outcomes:

The variance is:

Thus σ = √0.0136 ≈ 11.66%. The expected return is 6%, but the distribution indicates material loss exposure.

2.1 Common distributions

The normal distribution is defined by mean μ and standard deviation σ. It is mathematically convenient and often useful for aggregated measurement errors, but asset returns may exhibit skewness, fat tails and volatility clustering. A lognormal model is sometimes applied to positive asset prices because a continuously compounded return can be modeled as normal while the price remains positive. Neither assumption should be adopted merely for convenience.

The binomial distribution models the number of successes in n independent trials with constant success probability p:

Investment applications include simplified option trees, although real markets rarely satisfy all binomial assumptions exactly.

3. Sampling, Estimation and Statistical Inference

Investment decisions commonly rely on samples of historical returns rather than the full data-generating process. For observations x₁, x₂, ..., x_n, the sample mean is:

The unbiased sample variance is:

The divisor n − 1 corrects, under standard random-sampling assumptions, the tendency of a sample mean estimated from the same observations to understate population variance. The standard error of the sample mean is:

A large sample does not automatically produce a reliable estimate: dependence, structural change and non-representative observations can dominate sampling error.

Confidence intervals

When a sample is sufficiently large or the population is approximately normal, an approximate 95% confidence interval for a mean is:

When the population standard deviation is unknown and the sample is small under normality, a Student’s t critical value is more appropriate:

A confidence interval is a procedure with a long-run coverage interpretation. It does not mean that there is a 95% probability that a fixed, already-computed parameter lies inside this particular interval.

4. Dependence, Diversification and Regression

4.1 Covariance and correlation

For two return variables R_A and R_B, covariance is:

The sample covariance replaces the expectation with a sum and divides by n − 1. Correlation standardizes covariance:

Correlation lies between −1 and +1. It measures linear association, not causation, and a low historical correlation is not a guarantee of future diversification during a crisis.

For a two-asset portfolio with weights w_A and w_B, where w_A + w_B = 1, variance is:

Equivalently, the covariance term can be written 2w_Aw_Bρ_ABσ_Aσ_B. The cross term explains why portfolio risk can be lower than the weighted average of individual risks.

For N assets, matrix notation is compact:

Here w is the column vector of portfolio weights, w′ is its transpose, and Σ is the covariance matrix. The matrix must be dimensionally compatible and, for a valid variance model, positive semidefinite.

4.2 Regression

A simple linear regression models an outcome as:

where y_i is the dependent variable, x_i is the explanatory variable, α is the intercept, β is the slope and ε_i is the unexplained error. In finance, a market model may be written:

Here R_f is the risk-free return and R_M is market return. Beta measures sensitivity to market excess return; alpha is the average intercept conditional on the model and sample.

The ordinary-least-squares slope in a simple regression is:

and the intercept is α̂ = ȳ − β̂x̄. R² measures the fraction of sample variation in the dependent variable explained by the fitted linear relationship, but a high R² does not establish causality or forecastability.

Important assumptions include an appropriate functional form, meaningful observations, correctly measured variables, exogeneity of the error conditional on regressors, and sufficiently stable relationships. Financial time series may violate independence and constant-variance assumptions. Robust standard errors, rolling estimation, factor expansion and out-of-sample testing may be required.

5. Scenario Analysis and Simulation

5.1 Scenario analysis

Scenario analysis evaluates results under a small number of coherent states, such as recession, baseline and expansion. It is particularly useful when relationships are nonlinear or historical data contain no close precedent. A scenario should specify assumptions for rates, inflation, earnings, exchange rates, liquidity and correlations, rather than merely attach arbitrary labels to returns.

A decision table can include:

For a portfolio invested 60% in equity and 40% in bonds, the portfolio return in recession is 0.6(−20%) + 0.4(6%) = −10.4%. Scenario analysis is not a forecast probability exercise alone; it is also a stress-testing discipline. Extreme but plausible scenarios should be included even when assigned low or unspecified probabilities.

5.2 Monte Carlo simulation

Monte Carlo simulation repeatedly draws random values from specified distributions and propagates them through a valuation or portfolio model. A generic process is:

1. Define the question, horizon, cash-flow rules and constraints.
2. Select distributions and parameters for risk factors.
3. Specify dependencies, such as correlations or copulas.
4. Generate a random trial and calculate the resulting portfolio value or return.
5. Repeat many times and summarize the empirical distribution.
6. Test sensitivity to distributions, parameters, correlations and seed selection.

For a normally distributed one-period return, a simulation may use:

where j indexes trials. Simulated returns should not be confused with observed outcomes. The simulation is conditional on model specification; more trials reduce Monte Carlo sampling noise but do not correct a bad model.

6. Downside-Risk Measures

6.1 Semivariance and downside deviation

Variance penalizes upside and downside equally. Given a target return T, downside semivariance can be estimated as:

Downside deviation is √SV_T. Some conventions divide by the number of observations below the target rather than by n; the convention must be disclosed because results differ.

6.2 Value at risk

Value at risk (VaR) at confidence level c is a loss threshold not exceeded with probability c over a stated horizon, under a stated information set and model. If L denotes a positive loss, then:

For a normal return model with portfolio value V, mean return μ and standard deviation σ, a simplified loss VaR is:

where z_c is the standard-normal quantile. This formula assumes linear exposure, normal returns and a one-period horizon. If the result is negative, the assumed mean return exceeds the selected volatility threshold; practitioners may report a floor of zero or retain the signed result according to policy.

VaR does not describe the average loss beyond the threshold. Conditional VaR (CVaR), also called expected shortfall, does:

For continuous distributions, CVaR is the average of tail losses at or beyond VaR. It is generally more informative about extreme losses and has desirable aggregation properties under broad conditions. Historical, parametric and Monte Carlo methods can each estimate VaR or CVaR, but their outputs differ when tails are non-normal.

Worked example: parametric VaR

A £10 million portfolio has expected one-day return 0.04% and daily volatility 1.2%. At 95% confidence, z_0.95 ≈ 1.645. Under the normal, linear model:

This means the model estimates a one-day loss threshold of approximately £193,400 at 95% confidence. It does not mean that the maximum possible loss is £193,400, nor that the estimate is reliable during a regime change.

7. Illustrative Integrated Case: Northbridge Pension Fund

Background

Northbridge Pension Fund is considering a strategic allocation of 60% global equities and 40% investment-grade bonds. The investment committee must assess whether the allocation can meet a 5% nominal annual objective while limiting severe one-year losses. This is an illustrative case, not advice to a specific fund.

Problem and data

The committee has ten years of monthly returns, but the period contains unusually supportive monetary conditions. Estimated annual expected returns are 8% for equities and 3% for bonds; annual volatilities are 18% and 6%, respectively. The estimated correlation is 0.15. The committee also defines three forward-looking scenarios: recession, baseline and expansion.

Quantitative analysis

Using the portfolio-variance formula:

The weighted expected return is:

Using the simplified normal VaR formula and ignoring the mean for a conservative approximation, one-year 95% VaR is approximately:

The scenario analysis produces the following portfolio returns: −10.4% in recession, 6.6% in baseline and 11.2% in expansion. These results are not directly comparable with the normal VaR without aligning horizons and probability definitions, but they reveal an important issue: the recession scenario is less severe than the normal one-year 95% threshold, while a more severe market shock remains possible.

Approach and governance response

The committee should not approve the allocation solely from the historical covariance matrix. It should:

• conduct historical and hypothetical stress tests, including equity–bond correlation rising sharply;
• estimate volatility using alternative windows and, where justified, volatility models;
• perform a Monte Carlo analysis with fat-tailed and regime-sensitive alternatives;
• distinguish nominal from real returns and incorporate liabilities, inflation and cash-flow timing;
• define rebalancing, liquidity and leverage rules;
• validate data, code and model documentation independently;
• report both central estimates and uncertainty ranges.

Outcome and learning points

The allocation appears to meet the nominal expected-return objective, but the result depends heavily on expected returns, correlation and the assumption that the historical sample represents the future. The case demonstrates that diversification is conditional, VaR is model-dependent, and scenario analysis should complement—not replace—statistical estimation.

8. Limitations, Ethics and Professional Practice

Data limitations

Historical returns may contain survivorship bias, look-ahead bias, stale pricing, backfill bias and changes in market structure. Corporate actions, dividends, transaction costs, taxes and illiquidity must be treated consistently. A longer dataset is not automatically better if it combines incompatible regimes.

Model risk

Model risk arises from incorrect specifications, parameter uncertainty, coding errors and inappropriate use. Correlations can rise in crises; distributions can have fat tails; and estimated relationships can break when policy, technology or regulation changes. Analysts should distinguish parameter uncertainty—uncertainty about a model’s inputs—from model uncertainty—uncertainty about the model itself.

Governance and ethics

Investment professionals should disclose assumptions, data sources, estimation windows, missing data treatment, confidence levels and limitations. Risk measures must not be selected merely to produce an attractive presentation. Conflicts of interest, performance incentives and pressure to suppress adverse scenarios can create ethical hazards. Independent validation, version control, audit trails, segregation of duties and documented model-approval processes are appropriate controls.

Technology and AI

Spreadsheets, statistical software, cloud platforms and AI tools can improve speed and reproducibility, but they also introduce risks: opaque transformations, data leakage, hallucinated formulas, cyber compromise and overfitting. AI-generated analysis should be independently recalculated, tested on unseen data and reviewed by a qualified professional. Explainability is especially important when quantitative outputs affect beneficiaries, clients or regulated decisions.

Best-practice checklist

A defensible quantitative investment process should:

1. define the investment question and decision horizon;
2. match rates, returns and cash flows to consistent units and compounding conventions;
3. document data provenance and clean data without concealing inconvenient observations;
4. use economically plausible assumptions;
5. compare alternative models and conduct sensitivity analysis;
6. validate forecasts out of sample where possible;
7. stress nonlinear and tail outcomes;
8. communicate uncertainty rather than false precision;
9. monitor model performance after implementation; and
10. review assumptions when market regimes or portfolio mandates change.

Chapter Summary

Time value of money provides the basis for discounting cash flows, comparing projects and valuing securities. Present value, NPV, annuity and perpetuity methods require consistent rates, periods and cash-flow timing. Probability models express uncertainty, while expected return, variance and standard deviation summarize central tendency and dispersion. Sample statistics and confidence intervals support estimation but cannot eliminate bias or structural uncertainty.

Covariance and correlation explain the diversification effect, and the portfolio-variance equation shows why co-movement matters. Regression decomposes observed relationships into modeled and unexplained components, but statistical association is not causation and historical beta or alpha may not persist. Scenario analysis evaluates coherent alternative futures; Monte Carlo simulation generates distributions conditional on specified assumptions. Semivariance, VaR and CVaR focus on adverse outcomes, with CVaR providing information about losses beyond the VaR threshold. Across all methods, data quality, model validation, transparency, stress testing and professional judgment are essential.

Key Terms

Time value of money: The principle that cash flows at different dates have different economic values.

Present value: The value today of a future cash flow discounted at an appropriate rate.

Future value: The accumulated value of a current amount after compounding.

Net present value: Present value of inflows less the initial investment.

Internal rate of return: The discount rate that makes an investment’s NPV equal to zero.

Annuity: A series of equal cash flows at regular intervals.

Perpetuity: A stream of cash flows continuing indefinitely.

Expected return: The probability-weighted average of possible returns.

Variance: The probability-weighted squared dispersion around the mean.

Standard deviation: The square root of variance, expressed in the units of return.

Covariance: A measure of joint movement between two variables.

Correlation: Standardized covariance ranging from −1 to +1.

Beta: Estimated sensitivity of an asset’s return to a chosen factor or market return.

Alpha: The regression intercept, interpreted relative to the selected model.

Scenario analysis: Evaluation of outcomes under specified coherent states of the world.

Monte Carlo simulation: Repeated random sampling used to generate an outcome distribution.

Semivariance: Dispersion measure focused on observations below a target.

Value at risk: A specified quantile of loss over a stated horizon and confidence level.

Conditional value at risk: Average loss conditional on loss exceeding the VaR threshold.

Model risk: Risk that a model or its implementation is unsuitable, inaccurate or misused.

Review Questions

Short Answer Questions

1. Why must the rate and number of periods in a TVM calculation use consistent units?
2. Distinguish covariance from correlation.
3. What does the slope coefficient represent in a simple regression?
4. Why can standard deviation be insufficient as a measure of investment risk?
5. State one important limitation of Monte Carlo simulation.

Long Answer Questions

1. Explain the advantages and limitations of NPV and IRR for evaluating mutually exclusive investments.
2. Derive and interpret the two-asset portfolio-variance relationship, including the role of correlation.
3. Compare normal, lognormal and empirical approaches to modeling investment returns.
4. Explain how sampling error, structural change and data bias affect estimates of expected return and volatility.
5. Compare VaR and CVaR and discuss why a risk manager may report both.

Analytical / Application Questions

1. An asset has probabilities 0.30, 0.45 and 0.25 with returns −15%, 7% and 19%. Calculate its expected return and standard deviation.
2. A portfolio contains 70% of Asset A and 30% of Asset B. Their volatilities are 20% and 8%, and their correlation is −0.10. Calculate portfolio volatility and explain the diversification effect.
3. Design a stress-testing programme for a pension portfolio exposed to equity, interest-rate, inflation and currency risk. Specify scenarios, outputs, governance controls and limitations.

Multiple Choice Questions with Answers

Question 1

An amount of £10,000 is invested for two years at an annual compound rate of 5%. What is its approximate future value?

A. £10,500

B. £11,025

C. £11,500

D. £12,100

Correct Answer: B. £11,025

Question 2

Which measure focuses on the average loss beyond a specified VaR threshold?

A. Variance

B. Beta

C. Conditional value at risk

D. Correlation

Correct Answer: C. Conditional value at risk

Question 3

If two assets have a correlation of −1 and suitable portfolio weights, what may occur?

A. The portfolio must have twice the individual risk

B. Diversification can eliminate variance

C. Expected returns must be equal

D. Covariance is necessarily positive

Correct Answer: B. Diversification can eliminate variance

Question 4

For a discrete distribution, expected return is calculated as:

A. The largest possible return

B. The unweighted median return

C. The probability-weighted average of returns

D. The return minus its variance

Correct Answer: C. The probability-weighted average of returns

Question 5

In the regression y = α + βx + ε, what does β represent?

A. The unexplained error

B. The intercept

C. The sensitivity of y to x

D. The sample size

Correct Answer: C. The sensitivity of y to x

Question 6

A project has an NPV of £8,000 at the required return. What does this indicate, subject to the assumptions?

A. Its discounted inflows exceed its investment by £8,000

B. Its IRR is exactly zero

C. It has no risk

D. Its future cash flows are guaranteed

Correct Answer: A. Its discounted inflows exceed its investment by £8,000

Question 7

Which statement about a 95% confidence interval is most accurate?

A. There is a 95% probability that the fixed parameter moves daily

B. The procedure has 95% long-run coverage under its assumptions

C. It proves the sample is unbiased

D. It eliminates model uncertainty

Correct Answer: B. The procedure has 95% long-run coverage under its assumptions

Question 8

Which is a likely limitation of historical correlation estimates?

A. Correlation always measures causation

B. Correlation cannot be calculated from returns

C. Relationships may change during crises

D. Correlation must exceed one

Correct Answer: C. Relationships may change during crises

Question 9

A one-day portfolio VaR estimate is £200,000 at 95% confidence. Which interpretation is correct?

A. The maximum possible loss is £200,000

B. The average loss is exactly £200,000

C. Under the model, the selected loss quantile is £200,000

D. The portfolio cannot lose money on 95% of days

Correct Answer: C. Under the model, the selected loss quantile is £200,000

Question 10

What is the principal purpose of scenario analysis?

A. To guarantee the most probable forecast

B. To examine outcomes under coherent alternative assumptions

C. To replace all statistical estimation

D. To ensure returns are normally distributed

Correct Answer: B. To examine outcomes under coherent alternative assumptions