It has become readily apparent the Stock Trading Model requires a Stop Loss. A stop loss is a tool that traders use to throw in the towel on any particular trade once they've lost a predetermined amount of capital. In this case we will be using a predetermined percent of capital. Basically, it's to prevent anomalous price movements from wrecking the entire portfolio.
Before diving into the stop loss ramifications, let's do an overview of the historical statistical expectations of the Stock Trading Model it its' pre-stop loss form.
Below are the statistics for each Long, Short, and All combined trades used in testing this model:
A histogram of each Long and Short trade can be viewed below:
To summarize the left tails in each of the strategies I would say following: in the Long population there are 10 trades that lose between 20% and 25%, 6 trades that lose between 30% and 40%, and 2 trades that lose between 40% and 50%. In the Short population, which is much smaller, there are only 2 trades that lose more than 15%.
To my eye, there is a natural break at the 20% loss level, followed by trades that do not quite follow the normal profile of the rest of the population of trades. In other words we could put a stop loss at 20% and hopefully remove these anomalous rogue trades which could devastate our performance.
In doing this however, we will probably increase our losses on some trades that would naturally tick above 20% but end up well below that size of a loss. It's one of those laws of finance that sounds like something Newton would say about physics:
You cannot destroy risk without conceding some amount of potential reward. - David N. Dyer
Therefore we need to estimate how many of the trades in each bin below 20% would have been a 20% loss if the stop loss would have been implemented. Without rewriting the trade code and running through the simulations, I've just assumed that half of the trades that fell in the -15% to -20% bin and half of the trades in the -10% and -20% bin would've been stopped out. I think this is actually conservative, and instead of using the actual value of -20%, I've used the average of the -20% and -25% bin, which results in a value of -22.5%. This should account for some trades that blow through the -20% value and are stopped out below the mark.
This approximation results in 30.5 trades being stopped out in the Long population and 3.5 trades being stopped out in the Short population. You can see the modified histograms below:
If we rerun the approximated results through the Kelly Criterion, which you can read more about here and here, the expectations deteriorate only slightly. The overall f* from the Kelly formula deteriorates from 898% to 824%, suggesting the optimal expected returns have fallen somewhat. The added risk control of a stop loss far outweighs the very minor expected return deterioration.
We will therefore be implementing a 20% stop loss on each new trade initiated. If a trade is down over 20% from entry at the time of publishing (3:00-3:30 PM) the Trade Model will exit the position at the close.