Backtesting ETF portfolios

Backtesting ETF portfolios is a very important part of validating any investment strategy that uses them. At ETFMathGuy, we backtest our optimal portfolio construction strategy periodically. Doing so ensures that our quantitative methodology stays calibrated to the highest performing portfolios. Here, we discuss the key findings from this recent analysis.

Backtesting methodology

Our backtesting methodology follows the same approach we used in our previous backtesting analysis. The key distinction now is our time period begins in 2014 and runs through April of 2021. Also, we focused on one-month holding periods this time. Why? Based on our previous results, we found holding periods between 1-3 months had little impact on returns.

Backtesting ETF results over a longer-term

Firstly, the chart below shows the result of changing the duration of the sampling period on the out-of-sample returns. Note that there are two local maximums, with the first occurring and the 6-9 months, but a second more substantial maximum occurring at about 39 and 45 months.

However, when a risk-adjusted return is considered, we can improve this calibration. In the next figure, we show the annualized return divided by the annualized volatility. Thus, it’s clear that the 39 month sample period is superior with this measure for the moderate and aggressive portfolios. For the conservative portfolios, there is only a slight degradation in risk-adjusted return over these 7+ years of backtesting.

Backtesting ETF results over a shorter term

We also backtested our quantitative strategy over a shorter interval of the last 15 months, from January 2020, through April 2021. Ideally, our backtesting results over the long-term, shown above, should agree with this shorter time frame. And, in fact, they generally do.

Once again, with the slight exception of the conservative strategy, the 36-39 month sample size provided the largest annualized returns and risk-adjusted returns.

Key takeaways

• Backtesting provides an estimate on how our quantitative strategy would have performed based on historical time periods.
• The best calibration for the sample period occurs around 39 months based on both absolute return and risk-adjusted return.
• Longer-term and shorter-term backtesting provided similar calibration results.

Please leave this field empty

Oh hi there
It’s nice to meet you.

Sign up to receive expert commentary in your inbox, every month.

This field is required.

We don’t spam! Read our privacy policy for more info.

Please check your inbox or spam folder to confirm your subscription.