What can past returns tell us about future risk and returns?
Our analysis seeks to help determine whether there is a connection between recent returns (over varying look-back periods), and future returns. A positive connection might suggest that long-term alpha can be generated through momentum-oriented strategies, while a negative connection might suggest that mean-reversion is the dominant force.
In conducting the experiment, we looked at US equity prices for the most liquid 50% of publicly listed companies (around 4,000 companies, in total).
We then deployed an algorithm to rank stock returns by past performance over 5, 10, 15, 20 and 30 day periods, before categorising returns by decile (creating a 5x10 matrix of returns).
From here our algorithm takes over, plotting returns from each category over the next month. By rebalancing each month, we are provided with returns data that allows us to calculate:
Results
Over our sample period, we find that mean-reversion dominates momentum, both as a way to enhance returns and also minimise drawdown. Mean-Reversion also enhances the portfolio Sharpe, though we do see a weaker result from the worst performers than those within the 2nd to 4th deciles. This may perhaps suggest that assets that are sold off heavily, are sold for a reason, and to a level that is more in line with intrinsic value than just emotional trading.
While we hesitate to draw too many conclusions from this analysis, it would suggest that there may be opportunities for investors to outperform by shorting the best performing decile of stock over the last 5 and 10 days, while taking a long position in the worst performing 2nd, 3rd, 4th and 5th deciles over the last 5 and 10 days.
Applied to our sample period, this produced the following results (relative to sample mean):
Model attributes: Algorithm was built in Python and deployed via Quantopian using data from Quandl. 50 independent trials across 6,000 rebalancing periods. Supplementary analysis (factor and rules based weightings) conducted through Excel.
In conducting the experiment, we looked at US equity prices for the most liquid 50% of publicly listed companies (around 4,000 companies, in total).
We then deployed an algorithm to rank stock returns by past performance over 5, 10, 15, 20 and 30 day periods, before categorising returns by decile (creating a 5x10 matrix of returns).
From here our algorithm takes over, plotting returns from each category over the next month. By rebalancing each month, we are provided with returns data that allows us to calculate:
- Sharpe
- Beta
- Alpha
- Pre-Tax Returns
- Max Drawdown
Results
Over our sample period, we find that mean-reversion dominates momentum, both as a way to enhance returns and also minimise drawdown. Mean-Reversion also enhances the portfolio Sharpe, though we do see a weaker result from the worst performers than those within the 2nd to 4th deciles. This may perhaps suggest that assets that are sold off heavily, are sold for a reason, and to a level that is more in line with intrinsic value than just emotional trading.
While we hesitate to draw too many conclusions from this analysis, it would suggest that there may be opportunities for investors to outperform by shorting the best performing decile of stock over the last 5 and 10 days, while taking a long position in the worst performing 2nd, 3rd, 4th and 5th deciles over the last 5 and 10 days.
Applied to our sample period, this produced the following results (relative to sample mean):
- Sharpe
- Beta
- Alpha
- Pre-Tax Returns
- Max Drawdown
Model attributes: Algorithm was built in Python and deployed via Quantopian using data from Quandl. 50 independent trials across 6,000 rebalancing periods. Supplementary analysis (factor and rules based weightings) conducted through Excel.