Historical Simulation in Investment Analysis: Using Past Returns to Model Future Outcomes

Historical Simulation

Simulation forecasts possible outcomes for a system by testing different scenarios. The process involves four steps:

1. Setup: Define the goal and specify output variables (e.g., risk-adjusted return). Identify the input variables and the calculations that link them to the outputs.
2. Scenario generation: Create scenarios, ranging from simple user inputs to large, machine-generated datasets, often running thousands of simulations.
3. Evaluation: Apply the predefined calculations to each scenario using a calculation engine to produce output values.
4. Result analysis: Compile and analyze the outputs to conclude, such as constructing probability distributions for risk-adjusted returns to guide portfolio decisions.

Three common simulation approaches differ mainly in how they source scenarios:

• Historical simulation: Relies on actual past market movements to generate possible future results. It does not require any assumption about the shape of the return distribution.
• Bootstrap simulation: Resamples historical data with replacement to create synthetic datasets, introducing randomness and reducing overfitting.
• Monte Carlo simulation: Relies on random sampling based on distributional assumptions for input variables.

A robust simulation system can handle scenarios from any of these approaches via a central engine that processes a large number of scenarios to produce precise estimates.

Historical simulation uses past market data to project potential outcomes without assuming any specific return distribution. In finance, this means applying actual historical price movements to current portfolios. Its main advantage is simplicity—results are easy to explain, such as “the portfolio would gain INR300 million if last July’s market moves repeated.”

The method assumes that historical price changes reasonably approximate future risks. It naturally captures real-world patterns such as fat tails (leptokurtic distributions) and volatility clustering, in which high-volatility periods tend to follow one another. However, the key limitation is that past conditions may not hold during unprecedented events or regime shifts.

How to Implement It

A simulation engine can take scenarios from various sources. For historical simulation, you build scenarios by matching each input variable with its actual historical change on selected past dates.

The process follows the standard simulation framework:

• Setup – Identify key valuation and risk factors. For equities, use continuously compounded returns, \(r_t = \ln(\frac{P_t}{P_{t-1}})\). For bonds, model yield curve changes and discount cash flows using simulated yields, \(PV_B = \sum_{t=1}^{T} \frac{CF_t}{(1+y_t)^t}\).

• Scenario generation – Select historical dates and apply the observed changes to current factors.
• Evaluation – Process each scenario through the calculation engine.
• Analysis –  Compile results, e.g., computing value at risk (VaR) for a long-only portfolio by applying historical returns to current positions. This yields a profit/loss distribution without relying on distributional assumptions.

Historical simulation shines when modeling assets with non-normal returns – those exhibiting fat tails or skewness – since it captures extreme events that parametric models often miss. However, data quality is critical. Missing or unreliable data can arise from asynchronous global markets (different holiday schedules or unplanned closures), securities that didn’t exist throughout the historical period, or thinly traded instruments with sparse pricing. One solution is to use a proxy: a weighted average of similar securities with sufficient data to approximate the behavior of the missing instrument.

Using a bond’s own historical price changes can produce distorted results. A better approach is to look at how benchmark yields (for key maturities such as 1 month, 6 months, 1 year) have shifted, then estimate how those yield changes would affect the bond’s price. Alternatively, one can track the historical price movements of zero‑coupon bonds that match the bond’s remaining maturities. This avoids artifacts such as a soon-to-mature bond appearing volatile due to past price moves.

Strengths of Historical Simulation

One of the primary advantages is simplicity. Unlike parametric models that require complex distributional assumptions or parameter estimation, historical simulation directly applies past market movements to current portfolios. This makes it easy to implement and explain to stakeholders—for example, “this is what would happen if last year’s financial crisis repeated.”

A related strength is that it imposes no assumptions about return distributions. Real-world asset returns often exhibit fat tails, skewness, and volatility clustering. Historical simulation naturally captures these features without forcing the data into a normal or lognormal framework.

Because it uses real historical data, the method provides a realistic view of how a portfolio might behave under previously observed conditions, including extreme market events. This is particularly valuable for stress testing and regulatory capital calculations.

Additionally, historical simulation automatically preserves actual correlations among assets. There is no need to estimate a correlation matrix or assume stable relationships—the historical data inherently reflect the interdependencies that occurred.

Finally, it has a low computational cost compared to Monte Carlo methods. Since scenarios are drawn directly from historical records rather than randomly generated, no large-scale random number generation or repeated sampling is required.

Weaknesses of Historical Simulation

The most significant drawback is that it is inherently backward-looking. It assumes past market conditions are a reliable guide to the future, which may fail during unprecedented events, structural regime shifts, or entirely new types of risks. A crisis that has no historical precedent cannot be simulated using this method.

Limited scenario variety is another issue. The simulation is constrained by the specific historical window selected. If that window omits certain extreme events (e.g., a once-in-a-century crash), those events will never appear in the results, potentially leading to understated risk estimates.

Data quality and availability pose practical challenges. Historical data may contain gaps due to asynchronous global markets, securities that did not exist for the entire period, or thinly traded instruments with unreliable prices. Poor data integrity undermines the accuracy of the entire simulation.

Lastly, for certain asset classes, such as fixed income, a naive application of historical price changes can produce misleading results. A bond nearing maturity, for example, would show artificially high volatility if past price changes from when it had long duration are applied directly. This requires careful adjustments, such as using benchmark yield changes instead of raw bond prices.

Question

An analyst is using historical simulation to model a 10-year government bond with 2 years remaining to maturity. Rather than using historical price changes of the bond itself, the analyst models changes in benchmark yields for 3-month, 6-month, 1-year, and 2-year maturities. This approach is most likely intended to address:

1. The need to incorporate fat-tailed distributions into the simulation.
2. The assumption that past price changes will not reflect future price movements.
3. The problem is that the bond’s price volatility will decrease as it approaches maturity.

Solution

The correct answer is C.

As a bond approaches maturity, its price volatility naturally decreases. Using historical price changes from when the bond had a longer maturity would incorrectly suggest higher volatility. Focusing on benchmark yield changes for appropriate maturities and then inferring price impacts addresses this issue.

A is incorrect. Because historical simulation naturally captures fat tails from historical data, no special adjustment is needed.

B is incorrect. Because historical simulation explicitly assumes past price changes approximate future movements, this is a feature, not a problem, being addressed.