The Hidden Cost of Variance Targeting in GARCH Estimation

GARCH models are a staple of financial analysis, used to estimate volatility and forecast future returns. However, one technique that has gained popularity among practitioners is variance targeting. This method involves specifying the asymptotic variance, which can significantly impact the estimation results. But what exactly is variance targeting, and how does it affect GARCH estimates?

Variance targeting was born out of necessity in a time when computers were slower and optimization algorithms less sophisticated. It allows for quicker convergence of the optimization algorithm by fixing one parameter, the asymptotic variance, rather than estimating it from the data.

The Core Concept: Variance Targeting

The idea behind variance targeting is straightforward. In a GARCH(1,1) model, if you know alpha, beta, and the asymptotic variance (the value of the prediction at infinite horizon), then omega (the variance intercept) is determined. By specifying the asymptotic variance, you effectively eliminate one parameter from the estimation process.

This approach may seem appealing in terms of computational efficiency, but what are the implications? In a GARCH(1,1) model with t-distributed errors and 7 degrees of freedom, how does variance targeting affect the estimates?

The Mechanics: Data Points and Analysis

To investigate the impact of variance targeting, we generated series from a GARCH(1,1) model using t-distributed errors. We then estimated the parameters with and without variance targeting. The results are striking.

Figures 1 and 2 compare the half-life estimates with and without variance targeting. In both cases, the distribution of estimates is wider when no targeting is used. This suggests that variance targeting may lead to more precise estimates.

Portfolio Implications: Assets and Scenarios

But what does this mean for portfolios? We'll consider three scenarios: conservative, moderate, and aggressive approaches. Our assets are a mix of large-cap stocks (C, MS), quality bonds (QUAL), growth stocks (GS), and an aggregate bond fund (AGG).

In the conservative scenario, we may want to allocate more weight to QUAL and AGG. However, if variance targeting leads to more precise estimates, it could also result in overconfidence in the model's predictions.

Practical Implementation: Timing and Entry/Exit Strategies

So how should investors apply this knowledge? One approach is to use a combination of variance targeting and traditional estimation methods. This would allow for more robust estimates while minimizing the risk of overfitting.

Another consideration is timing. If variance targeting leads to quicker convergence, it may also result in earlier entry into the market. However, this could be counterproductive if the model's predictions are overly optimistic.

Actionable Conclusion: Synthesizing Key Insights

In conclusion, variance targeting can have a significant impact on GARCH estimates. While it may lead to more precise estimates and quicker convergence, it also introduces new risks. Investors should approach this technique with caution, considering both the benefits and drawbacks of variance targeting in their portfolio management strategies.