Volatility Optimization

Introduction

This optimization was started because I noticed that there are large time periods of successive losing trades.  Based on my market experience, I surmised that this momentum strategy is not profitable during these times because the momentum was so weak that the likelihood of a breakout or breakdown was low.  Thus, these time periods were characterized by successive trades where it hit the stop.  This can happen for days at a time, or over hours, but I needed a parameter for the strategy to avoid or at least minimize the damage.  Below is a sample stock with strategy performance over 10 trading days.

Minimum baseline volatilities were implemented in Test 2A and Test 2B with promising results.

Unmodified Strategy

Test 2A

Test 2B

Analysis

Within the test frame, increasing historical volatilities have tended to increase the percentage profitability of this momentum strategy.  However, this data is inconsistent as to the percentage of historical volatility which results in a higher percentage profitability for the signal, as some stocks require higher amounts of volatility for a corresponding increase in percentage profitability.  For example, AAPL's selectivity went 0.15% up for a 10.6% increase in win rate, and GS's selectivity went 0.65% up for a 11.8% increase in win rate; leading to ratios of 1.42% and 5.51% respectively.  These ratios can be understood as the "Increase in Percentage Profitability per Unit % of Historical Volatility."

In real-world application of this form of selectivity, results are likely to be an improvement over the unaltered strategy, but the minimum volatility after which trades are more likely to be profitable is not predictable at this time or presumably stable.

Since this was entry selectivity, the gross change in profitability of the strategy can be estimated by the product of the difference in losing/winning trades with the average losing/winning trade.  The resulting change in gross profit across multiple stocks is inconclusive as to the effectiveness of a volatility minimum in increasing the realized gross profit. 

The metric mostly reduces the frequency of losing trades more than losing trades (in 6/8 cases), and the change in gross profitability is mostly beneficial (in 5/8 cases), so the metric is passable in terms of improving selectivity, but still not preferable in terms of metric strength.

Market exposure has gone significantly down across the board, and since the average winning trade is stable, this shows the volatility metric is effective at taking an entry only when the probability for a profitable trade is present. 

Improvement in the percentage profitability of the metric is also seen in the average number of bars between winning and between losing trades, with the average number of bars between losing trades increasing more rapidly than the average number of bars between winning trades.

Conclusion

This strategy can be viewed as a "momentum following" strategy, since it seeks to enter breakouts/breakdowns, and success is more likely within a period of historical volatility.  The metric, as applied, does increase the percentage profitability, as well as reducing the market exposure.  However, this metric does not yet result in a clear effective change in gross profit.  This is largely because although the frequency of negative trades is reduced more than positive trades, in some cases the average negative trades are increased more than the average positive trades, in some cases the average positive trade went down while the average negative trade went up.  As a result, the effective gross profitability change of the metric over the null hypothesis is inconclusive. 

In conclusion, the volatility metric reduces churning, but does not effectively select for situations when the percentage profitability is high relative to its position.  Further investigation will be made into other metrics, which hopefully will increase the percentage profitability by a dramatic amount.  It is quite possible that the metric is subject to "false breakouts", which means that further investigations will progress towards quantifying anticipated momentum strength/weakness.  Assuming a suitable metric is found, further research will be done into a stop-loss strategy that will further improve the consistency of the strategy, and "probability metrics" which can be used in higher-probability situations to increase the size of the trade.