Understanding Stochastic Methods for Simulations in SAS Risk Engine

[#1]

Introduction

Uncertainty is at the heart of risk modelling. Economic indicators, market variables, and financial instruments rarely follow a perfectly predictable path. Their future values are influenced by both systematic trends and random movements.

Stochastic processes provide a mathematical framework for representing this uncertainty. By specifying how a variable evolves over time, stochastic models allow simulations to generate multiple possible future scenarios rather than relying on a single deterministic forecast.

In SAS Risk Engine, a stochastic process specification table is required as an input to facilitate Monte Carlo simulation. The table defines the stochastic method and its associated parameters for each risk factor. Once configured, these specifications help the simulation generate realistic paths for the selected variables.

What Is a Stochastic Process?

For detailed information on SAS Risk Engine,

please refer to the following:

SAS Risk Engine Help Documentation

A stochastic process describes how a variable changes over time when its future behavior contains an element of randomness.

A simplified representation is:

Future Value = Systematic Component + Random Component

Different stochastic methods define these components differently. Some emphasize a long-term trend, while others assume that a variable tends to return toward a long-term average.

The choice of method is therefore important because it determines the characteristics of the simulated paths.

Common Stochastic Methods

Several stochastic methods are commonly used in financial and risk simulations.

1. Geometric Brownian Motion

Geometric Brownian Motion (GBM) models changes that are proportional to the current value of a variable. It is commonly associated with processes where percentage changes are more meaningful than absolute changes.

The process can be represented as:

dS = μS dt + σS dW

where:

• μ = drift
• σ = volatility
• dW = random Brownian shock
• S = current value of the variable

The presence of S in both the drift and volatility components means that the magnitude of movements changes with the level of the variable.

GBM is therefore useful when modeling variables that exhibit proportional growth and volatility.

2. Brownian Motion 

Brownian Motion provides a simpler additive stochastic process:

dX = μ dt + σ dW

Here, the random movement is not directly multiplied by the current value of the variable.

The two key parameters are:

• Drift (μ): the systematic directional movement
• Sigma (σ): the magnitude of random variation

This method can be useful when an additive stochastic process is appropriate and mean reversion is not required.

3. Cox-Ingersoll-Ross

 

The Cox-Ingersoll-Ross (CIR) model introduces mean reversion and level-dependent volatility.

Its process can be represented as:

dX = κ(θ − X)dt + σ√X dW

where:

• κ = mean-reversion rate
• θ = long-term mean
• σ = volatility
• X = current value

The term κ(θ − X) pulls the variable toward its long-term mean.

An important characteristic of CIR is that volatility depends on the current level through √X. This makes it different from simpler mean-reverting models.

4. Vasicek

The Vasicek model is another mean-reverting process:

dX = κ(θ − X)dt + σ dW

It uses the same basic mean-reversion concept as CIR, but the volatility component is constant rather than dependent on the current value.

The key parameters are:

• Mean Reversion Rate (κ)
• Mean (θ)
• Sigma (σ)

Vasicek is therefore a relatively simple way to represent a variable that fluctuates around a long-term level.

A Stochastic Process Specification Example

Consider the following configuration in a SAS Risk Engine stochastic process specification table:

Method	Risk Factor	Parameter	Value
GeometricBrownianMotion	HPI	Drift	0.02
GeometricBrownianMotion	HPI	Sigma	0.10
BrownianMotion	VIX	Drift	0.04
BrownianMotion	VIX	Sigma	0.20
CoxIngersollRoss	DOW	MeanReversionRate	0.01
CoxIngersollRoss	DOW	Mean	27,850
CoxIngersollRoss	DOW	Sigma	0.25
Vasicek	SP500	MeanReversionRate	0.02
Vasicek	SP500	Mean	3,035
Vasicek	SP500	Sigma	0.15

This example illustrates how different risk factors can use different stochastic processes within the same simulation framework.

For HPI or housing price index, Geometric Brownian Motion is configured with a drift of 2% and sigma of 10%.

For VIX or volatility index Brownian Motion uses a drift of 4% and sigma of 20%.

For DOW or Dow Jones industrial average,  the CIR model introduces a long-term mean of 27,850, a mean-reversion rate of 0.01, and sigma of 25%.

For SP500 or Standard & Poor's 500 index,  the Vasicek model uses a long-term mean of 3,035, a mean-reversion rate of 0.02, and sigma of 15%.

The example demonstrates an important modeling principle: different risk factors do not necessarily need to follow the same stochastic process.

Understanding the Key Parameters

Although each method has its own characteristics, several parameters appear repeatedly.

Drift (μ)

Represents the underlying directional tendency of a process.

Sigma (σ)

Represents the magnitude of random fluctuations. Higher sigma generally produces greater dispersion across simulated paths.

Mean (θ)

Defines the long-term target level for a mean-reverting process.

Mean Reversion Rate (κ)

Controls how strongly, or how quickly, the process moves toward its long-term mean.

These parameters work together to determine the behavior of the simulated scenarios.

Why Method Selection Matters

Selecting a stochastic method is more than a configuration decision. It is a modeling assumption about how a risk factor behaves.

For example:

• GBM is appropriate when proportional changes and growth are important.
• Brownian Motion is useful for additive random movement.
• CIR is useful when mean reversion and level-dependent volatility are important.
• Vasicek is useful when mean reversion with constant volatility is appropriate.

Choosing the wrong process can produce simulated scenarios that do not adequately represent the characteristics of the underlying risk factor.

From Specifications to Monte Carlo Simulation

Once the stochastic process and its parameters are provided through the stochastic process specification table, SAS Risk Engine can use these definitions as inputs to Monte Carlo simulation.

The simulation applies the specified stochastic processes to generate multiple possible paths for the risk factors. These scenarios can subsequently support activities such as:

• Risk measurement
• Scenario analysis
• Stress testing
• Forecasting
• Portfolio analysis

The strength of the approach comes from generating a distribution of possible outcomes, rather than relying on a single predicted value.

Here is an example of CAS Risk Action set using the specification table for simulation purposes.

B24P1-1024x656.png

 Illustration of a CAS risk action set for generating simulated states.

This sample code uses the riskSim.genSimulation action in SAS Risk Engine generates Monte Carlo simulation scenarios using the current market data (currMktTable), historical data (histTable), and stochastic process specifications (stochasticProcessSpecTable) as inputs. The generated simulation environment is stored in envOut, while the simulated risk-factor states are written to statesOut. The simulation is configured with 4 horizons (nHorizons=4) and 101 draws (nDraws=101), using 54321 as the random seed (seed=54321) for reproducibility, starting from 10-Jan-2020 (asOfDate) with a daily interval (interval='DAY'). The TWISTER random-number generator (generator='TWISTER') is used along with antithetic variates (useAntithetic='TRUE') to improve simulation efficiency.

Conclusion

Stochastic methods provide the foundation for representing uncertainty in simulation-based risk modeling. Methods such as Geometric Brownian Motion, Brownian Motion, Cox-Ingersoll-Ross, and Vasicek offer different ways of describing growth, randomness, volatility, and mean reversion.

In SAS Risk Engine, the stochastic process specification table provides the configuration needed to use these methods as inputs for Monte Carlo simulation. Understanding what each method represents—and how parameters such as μ, σ, κ, and θ influence the resulting paths—helps modelers make informed decisions when building stochastic simulations.

Find more articles from SAS Global Enablement and Learning here.