As I mentioned in an earlier post here at DDI, for the past few months, I’ve been working on an algorithmic options trading system. I described it as follows:
But, one rabbit hole tends to lead to another. The same filtering process allowed me to identify option candidates for short-term purchase and sale. By exclusively purchasing call and/or put options on these stocks, I could (on a projected basis) do very well because the leverage available in options far exceeds the leverage available in purchasing equities. Once again, I would be initially focused on stock price appreciation. If there were stocks that satisfied the screeners, I could then buy call and/or put options on them (depending on certain trend data). In fact, I had to buy options on all of them because my analytics required the purchase of “all” to make money from the performance of “some.” When I first published this approach, I said that I would trade in a simulator for a few months and then start to put real money to work. Then, I’d get back to my colleagues here at DDI and update everyone.
Here’s my interim report.
In an effort to reduce the number of purchase candidates, I tried to identify a set of variables that would lead to a set of candidates with a high probability of success (i.e., a profit on the sale of the option). I identified thirty-one variables that might have some predictive value (e.g., RSI, MACD, the Greeks, etc.). Notably, these thirty-one variables could be put together in over one billion combinations. Working with a statistician, I identified fourteen variables that taken together, did in fact support a higher probability of identifying successful candidates. But the probability was barely higher than 55% — which was not much better than a coin toss unless I was doing thousands of trades which would be beyond my capabilities and financial resources. So, I abandoned that effort and went back to my original thesis: buy all candidates and trade them.
But, I did not want to actually trade them – at least not the way that most people think of trading. I wanted to set a stop loss and a stop gain, which would allow me to do nothing after the initial purchase. My goal was to take the subjective assessment out of the equation – to be exclusively data-driven.
One of the great challenges of using stop losses is the possibility of triggering a stop loss on a position that would ultimately be successful if I were willing to hold it for a while. I had hoped to find a loss percentage that was routinely associated with a rebound in value. For example, did the data show that an option that fell no more than 5% ultimately became a winner in 75% (or more) of such transactions? Unfortunately, the data did not support that hypothesis. In fact, it was impossible to say that a particular loss level predictably led to winning positions (even marginally winning positions).
Given my first fundamental principle of buying all candidates that fit my purchase criteria), I ended up establishing a second fundamental principle: sell all losers when their value drops by a percentage that I would arbitrarily find acceptable. (More on this point in a moment.)
Using data to find a stop gain level was marginally more successful. It was clear that, over the fifteen-year period that was examined, at least 50% of all winning options gained between 1% and 40% in value (over three or five-day holding periods). But, all those gains occurred at unpredictable moments! Moreover, there was a meaningful percentage of options that ended up being worth far more than a 40% increase in value. Again, those increases in value occurred at unpredictable moments and I did not want to sit and watch a screen all day waiting for such values to emerge. I wanted to “set it and forget it.” The question was “when to sell?” Put another way, was there a level of profitability that was not (on a statistical basis) worthwhile to pursue?
If there is such a statistical methodology, then I couldn’t figure it out. However, since I had already “capped” my losses (by selling any loser during the holding period at a predetermined percentage loss), the only remaining decision that I had to make was when to sell winners. It was certainly true that 50% of all gains were between 1% and 40%. So, it might have been reasonable to have a stop gain set at 50% and sell everything at the end of trading period (i.e., sell all positions where the losses were less than 5% and gains were less than 50%). On the other hand, holding off and setting a stop gain at 100% was clearly more profitable (i.e., by giving me the opportunity to capture those larger gains between 50% and 100% at the end of the holding period although it made me subject to price declines during the holding period (i.e., an option that went up in value could also go down in the value during the holding period).
The actual length of the holding period also had an impact but, again, I made a subjective judgment: the additional profitability associated with a slightly longer holding period was not subjectively worth it for me. As I have said many times, I will always “swing” for averages and never for home runs.
I want to be clear: my back-testing did not prove or even suggest that my trading would be profitable each and every month. Rather, it simply showed that, on average, measured annually and over fifteen years, it was possible to set up a structure that had consistently demonstrated profitability of varying amounts. However, it integrated three data-guided but ultimately subjective decisions – the stop loss level, the stop gain level and the holding period.
Can I lose money? Absolutely. In fact, my research demonstrates that there were always losing months in every year over the last fifteen years (although the monthly loses were never more than 5% of invested capital plus transaction costs – i.e., the stop loss amount). The back-testing also showed that, over 180 trading cycles (i.e., the first three days of each month over the last 15 years), I never “lost” money in any trading year and that net profits on invested funds were approximately 9% per month on average. In the real world, I’d be thrilled with 5% per month. We will see.
My goal as a trader is to take the emotion out of the process and live a life unfettered to a screen. I want to look at my computer on Monday morning and again on Wednesday afternoon. So, these are the rules that I will be using:
- Buy an equal dollar amount of call and put options on all NASDAQ stocks that have increased in value over the preceding 42 trading days by a defined (and substantial) percentage.
- Hold those options for three days.
- During the holding period, sell all options triggered by a small, fixed percentage loss.
- During the holding period, sell all options triggered by a substantially larger, fixed percentage gain.
- At the end of the holding period, sell all previously unsold options.
- Take the rest of the week off.
This month, I intend to expand my back-testing to include at least four trading cycles per month for each of the last fifteen years. I’d be surprised if the results are materially different but there is no substitute for doing the research.
I am still committed to being a data-driven investor. I am still committed to using price as the primary metric of performance when I buy a position. And, I am still committed to identifying and buying bundles of positions that meet various statistical metrics independent of the standard metrics that are traditionally used to value stocks (e.g., ROE, E/P, PEG, etc.). I have learned that patterns do emerge in stock and options trading data; but, as I expected, they do not emerge predictably for any one stock or option over short periods of time. Therefore, for me, the better (i.e., more reliably profitable) approach is to buy the universe of positions that meet certain criteria and trade based on the chosen parameters.
From time to time, I will write a follow-up post about my progress and hopefully, DDI will allow me to publish it here.