/

Volatility Forecasting in Alternative Investment Funds Management

2171 views
5 mins read

The first half of 2022 has been a half to forget for AIFM.

It was the worst half of the year for the S&P 500 index since 1970, and there was nowhere else to hide for investors as the Bloomberg US Aggregate bond index fell by more than 10% in the same period.

Several macro factors had driven the sell-off.

They include the highest inflation readings since the 1970s, global rate hikes, recession fears, a global pandemic, supply chain disruptions, and increased geopolitical tensions.

As these factors remain unresolved, AIFM must be prepared for an environment of heightened volatility as the world learns something new about these global developments each day.

Volatility prediction models

Against this backdrop of elevated macro and market risks, a powerful tool that AIFM can incorporate into their risk management toolkits is volatility analysis and modelling.

Volatility refers to the dispersion of financial asset returns over time — the higher the volatility, the riskier the asset.

Due to specific statistical characteristics, we can better model volatility as compared to most market variables which are unpredictable.

The two traditional and most common volatility prediction methods are:

  •  ARCH: Autoregressive Conditional Heteroskedasticity (first introduced by Robert Engle in 1982)
  • GARCH: Generalized Autoregressive Conditional Heteroskedasticity (first introduced by Tim Bollerslev in 1986)

In statistics, an autoregressive model specifies that the output variable depends linearly on its previous values and imperfectly predictable terms.

An autoregressive model uses past values/behaviour to predict future values/behaviour as the past and present are correlated.

In financial markets-speak, periods of high volatility tend to be followed by high volatility, and vice versa.

As for heteroscedasticity, the term refers to a state where the standard deviations of a predicted variable are non-constant.

The GARCH model is an extension of the ARCH model that adds to the autoregression component of the ARCH model with a moving average component.

GARCH models enable modelling a broader range of behaviour and more persistent volatility.

For the ARCH model, the main disadvantage is that both positive and negative shocks have the same impact on the model’s volatility prediction.

However, in reality, positive and negative shocks tend to have a different impact on asset prices due to investor psychology (e.g., loss aversion).

As for the GARCH model, the primary disadvantage is that the model does not respond asymmetrically to falling and rising levels of volatility.

Therefore, a modified model, the GJR-GARCH, has been proposed to remedy the issue.

How AIFM use volatility prediction models

There are many applications for volatility prediction models.

Risk managers can incorporate these prediction models into the stress testing process.

Under the AIFMD, stress testing is obligatory as part of the liquidity management requirements (Article 16(1) of the AIFMD)) in market risk computation – using a Value-at-Risk (VaR) approach or an expected shortfall (ES) approach.

Besides stress testing, portfolio managers and traders can also use the ARCH and GARCH models to model the long-run standard deviations of the markets and contracts they trade, which help guide pricing and hedging, market making, market timing, and portfolio construction.

Besides modelling idiosyncratic volatilities, AIFMs can also leverage multivariate ARCH and GARCH models to forecast correlations and covariances across a portfolio of assets.

AIFM can automate the volatility prediction models and process as part of the routine risk management process and further improve the out-of-sample forecast with artificial intelligence and machine learning.

The ARCH and GARCH models have become essential forecasting tools in finance and economics, as many asset prices (e.g., equity, bonds, commodities) and economic metrics (GDP, inflation rate, unemployment rate) are conditional heteroskedastic.

The models have also expanded to other industries, such as healthcare.

Compared to the ARCH model, the GARCH model is more economical and appropriate when the data exhibits heteroskedasticity and volatility clustering.

The GARCH model also provides a more real-world context when modelling the prices and rates of financial instruments.

Although the GARCH model is not perfect, several GARCH extensions help to negate these issues.

Ultimately, as volatility is the purest measure of risk in financial markets and is more forecastable than returns, AIFM must leverage these volatility prediction tools to guide risk management and asset allocation in today’s increasingly uncertain world in the specific context of each AIF under management.

By Maryna Chernenko, Managing Director of UFG Capital