Performance of 52 Week Momentum in Factor Analysis

Earlier I looked at PE ratios as a factor (here). Another well known and simple factor supported by academic research is to buy companies that have performed the best over the last 52 weeks or as part of a long/short strategy.

I performed a factor backtest on 52 week total returns using the parameters below and took a closer look at the statistical properties of the strategy.

Parameters

Period of Test: 1999-01-02 to 2013-12-31
Re-balancing Frequency: Monthly
Ranking Method: Percentile into 5 buckets (0 to 20% (Lowest 52 week return), 20% to 40%, etc)
Stock Universe: Russell 3000
Factor: Trailing 52 Week Returns

Additionally, the long/short strategy is created by going long the stocks with highest 52 week returns (80 to 100) and going short the lowest 52 week returns (0 to 20).

Annualized Return Statistics

As expected, overall the top quantile of companies continued to outperform (12.05% annualized vs 3.21% annualized for the Russell 3000). Surprisingly, the overall long/short strategy did not perform very well as it’s performance was very weak coming off the market bottoms in 2003 and 2009, eliminating any positive returns generated in other years.

Strategy Annualized Return Annualized Std Dev Annualized Sharpe (Rf=0%)
Russell 3000 3.21 17.83 0.1801
0 to 20 1.47 39.11 0.0376
20 to 40 7.77 25.38 0.3063
40 to 60 9.39 19.51 0.4813
60 to 80 10.58 18.02 0.5869
80 to 100 12.05 23.65 0.5093
Long/Short 0.68 28.64 0.0237
52 Week Return Backtest: Annualized Return Barplot

Cumulative Return

On a cumulative and rolling basis the top performers outperform other quantile’s but the weakest quantile outperforms significantly coming off the bottoms of a bear market. This weak quantile out-performance post bear market can lead to a very poor period for a 52 week return long/short strategy.

52 Week Return Backtest: Historical Cummulative Returns
52 Week Return Backtest: Rolling 12 Month Performance Chart

Relative Performance Statistics

Where the long/short strategy does outperform is during a market downturn, as seen in strong performances by the strategy from 2000 to 2003 and 2008.

52 Week Return Backtest: Rolling 12 Month Relative Performance Chart
Risk Metric 0 to 20 to Russell 3000 20 to 40 to Russell 3000 40 to 60 to Russell 3000 60 to 80 to Russell 3000 80 to 100 to Russell 3000 Long/Short to Russell 3000
Alpha (%) 0.0000 0.3900 0.5100 0.6100 0.7400 0.7400
Beta 1.8450 1.2821 1.0081 0.9363 1.1346 -0.7104
Beta+ 2.1713 1.4781 1.0423 0.8677 0.8849 -1.2864
Beta- 1.4797 1.2342 1.0519 0.9954 1.2256 -0.2541
R-squared 0.7071 0.8111 0.8485 0.8579 0.7315 0.1955
Annualized Alpha (%) 0.0300 4.7500 6.3300 7.5300 9.2300 9.2100
Correlation 0.8409 0.9006 0.9211 0.9263 0.8553 -0.4422
Information Ratio -0.0670 0.3763 0.8135 1.0696 0.7075 -0.0635
52 Week Return Backtest: Rolling 12 Month Alpha/Beta Regression Chart
52 Week Return Backtest: Annualized Return_and_Risk Scatterplot


Correlations Statistics

This table shows the correlations of each of the strategies to the Russell 3000, including there p-value and confidence intervals. The long/short strategy shows a statistically significant negative return to the overall Russell 3000.

Strategy Correlation p-value Lower CI Upper CI
0 to 20 to Russell 0.84 0 0.79 0.88
20 to 40 to Russell 0.90 0 0.87 0.92
40 to 60 to Russell 0.92 0 0.90 0.94
60 to 80 to Russell 0.93 0 0.90 0.94
80 to 100 to Russell 0.86 0 0.81 0.89
Long/Short to Russell -0.44 0 -0.55 -0.32


Risk Metrics Statistics

The long/short strategy exhibits a long negative tail of monthly returns.

52 Week Return Backtest: Annualized Return Barplot

Looking at the overall risk metrics for the various strategies the 60 to 80% quantile shows the least volatile monthly returns. While again the long/short strategy has significant downside and max drawdown risk characteristics due to it’s performance in 2003 and 2009.

Risk Metric Russell 3000 0 to 20 20 to 40 40 to 60 60 to 80 80 to 100 Long/Short
Semi Deviation 3.94 7.48 5.25 4.24 4.04 5.25 6.81
Gain Deviation 2.83 8.95 5.04 3.38 2.82 3.73 4.38
Loss Deviation 4.02 7.12 5.37 4.46 4.24 5.27 8.96
Downside Deviation (MAR=10%) 4.16 7.53 5.22 4.20 3.97 5.07 6.95
Downside Deviation (Rf=0%) 3.75 7.11 4.83 3.83 3.60 4.69 6.65
Downside Deviation (0%) 3.75 7.11 4.83 3.83 3.60 4.69 6.65
Maximum Drawdown 51.51 83.00 60.38 52.01 49.21 54.70 79.44
Historical VaR (95%) -8.38 -16.66 -9.49 -7.31 -8.01 -10.70 -12.53
Historical ES (95%) -12.19 -22.49 -15.43 -12.24 -11.57 -15.44 -24.74
Modified VaR (95%) -8.78 -14.25 -10.82 -8.92 -8.51 -11.14 -15.74
Modified ES (95%) -14.91 -14.83 -19.11 -19.67 -16.95 -17.42 -34.98


Overall Remarks

  • Overall, those companies which performed the best over the last year are likely to continue to outperform over the next month.
  • During a market correction a long/short strategy based on 1 year returns is negatively correlated to the market, providing a hedge and will likely outperform.
  • As you come out of a bear market, the best performers in the previous year will under-perform (significantly in ‘09 case) those that performed the weakest. Additionally, continuing to allocate capital to a long/short 1 year return strategy immediately following a bear market can be very damaging to your portfolio.

By Jon Eickmeier | January 2, 2014 | analysis markets quantitative factor backtest

Comments Section

comments powered by Disqus