The FuseTrader reversion stock trading model is a set of rules-based algorithms developed using two disparate sets of historical time series data tracking the S&P 100 ($OEX) constituents through multiple market cycles. The first set was from 2006-2012 and the second set was 2014-2019. These sets were arbitrarily chosen albeit with the knowledge that the first set captured a recession and bear market and the second set was broadly in an uptrend. There were a few fundamental research ideas for the rationale discussed so far:
- The S&P 100 is comprised of the largest companies in the US markets; if a trading model can identify extremes in pricing on the largest companies and successfully arbitrage these pricing events, it should perform even better on companies with relatively higher price variation.
- Using two data sets allows the researcher to test ideas on one set at a time and then confirm the idea on the second set; all in an attempt not to 'curve fit', which is one of the common pitfalls in backtesting.
- Testing rules based systematic strategies on individual companies within different types of market conditions should be a primary condition of a robust trading model which should be useful in future, regardless of the sort of market conditions the future may hold.
Types of Trading Systems
In general, there are two types of systematic, rules-based trading systems:
Trend Following - these systems attempt to identify trends in pricing so that the model initiates a trade and rides the trend toward profitability. These systems typically work well in volatile markets but can get chopped up or whipsawed during relatively stable market environments. Trend following strategies typically contain less than a 50% win rate but have a win ratio greater than 1. This means most of the trades do not work, but hopefully the winning trades are large enough to compensate for all of the losers.
Overall these systems can work on a systematic basis, but they are generally exceedingly frustrating to follow and stick to. Most people will get excited with the winners and then stop following the system during a losing streak, typically bowing-out right before the next winning trade resumed the user to profitability. Therefore in a quest for the ideal trading system, you have to be confident that you'll be able to stick with it.
Reversion Systems - reversion strategies attempt to identify extremes in price wherein the probabilities of a price trend exhausting are such that the trend reversing becomes highly likely. These systems tend to have high win rates, much higher than 50%, but when they encounter a loss it is typically larger than the average win. The risk inherent in these systems is that you enter too soon and a runaway price trend catches the trader off-guard and a large loss takes place.
Due to the higher win rate, these systems tend to work better than trend following systems primarily because they are much more usable. The ease in usability allows the user to stick to these systems and if they are developed well they can significantly outperform what the average trader is able to accomplish on a discretionary basis alone.
The FuseTrader model utilizes numerous reversion based algorithms in order to identify multiple measures of price extremes and exhaustion. Overall we should expect actual long-run results to fall in-line with the following historical results:
"The central problem for gamblers is to find positive expectation bets."
Based on these statistics, I believe we have solved the central problem for "gambling" in the stock market.
The historical trades result in the following histogram for each Long and Short positions:
As you can see, we have geared the Trading Model to make more Long bets than Short bets however, each on it's own is statistically profitable.
This topic is discussed in further detail on the Portfolio Process write-up, but I have provided the Kelly results for each Long and Short historical trades below.
Given the following formula, combined with the historical trade statistics provided in the table above...
f* = p/a - q/b
- p = win probability
- q = loss probability or (1 - p)
- a = non-win loss
- b = win profit
- f* = amount of capital to deploy on any single bet
The results are as follows:
These results demonstrate that each of the Long and Short and strategies, as well as the combined approach is expected to be profitable, so much so that the Kelly formula suggests adding varying degrees of leverage for optimal returns. This formula is how I judge the expectations of a trading system, and the higher the f* the better.
The model tracks every S&P 500 constituent, every Nasdaq 100 constituent, and numerous others that I have a personal or professional interest in which have also been included in the suite of tickers that are modeled. I have also included index etf's for every major index as well as market sector.
As we progress with publishing the trades from this Trading Model and trading the signals ourselves, any revisions or tweaks to the methodology or additional research completed will be published in a blog format, located below: