A little history for the uninitiated.
About six months ago we launched a new trading strategy YK that was a radical departure from our original MarketSci strategies. YK is adaptive, meaning it was built to “learn” from the markets rather than using more traditional static rules.
In our first 6 months of trading, in some pretty ugly markets, YK did amazingly well, returning an annualized pace of 150.8% with a peak month-end drawdown of -2.1%.
Then came the October stock market slide. As I’ve discussed recently, this month has been unprecedented. Never before has the market come even close to losing this much this fast in consecutive down closing days without a technical bounce. YK wasn’t prepared for such a significant historical divergence and to put it bluntly, crashed and burned.
The strategy has lost over 30% this month. That’s not a typo.
On a personal note, as I wrote in Psychology of the Drawdown, as a trader with thick skin born of many years going toe to toe with these markets, I take it all in stride. Hell, we’ve still outperformed the market by a healthy margin over the last 6+ months. But as a guy who folks rely on for market wisdom, I carry a pain in my soul that nothing I do from now moving forward can ever hope to assuage.
What Went Wrong
This part all sounds like common-sense (in hindsight).
The YK Strategy was built assuming that (a) markets act in accordance with history (all technical strategies make this assumption), but that (b) history is constantly evolving and so the strategy must evolve with it (this is the concept that I think makes YK so unique).
YK also implicitly assumed that when the market diverged from history, that divergence would be brief enough and painless enough that the strategy could ride it out. This month has proven that assumption wrongheaded – the market can stay irrational far longer than you or I can remain solvent.
It’s important to note that I don’t think the “adapting” portion of the strategy failed. The market has never moved this way in 60 years of history; there’s no history to “adapt” from. What failed was the strategy’s inability to recognize that the market could act in a way that’s so much different than what’s been seen before.
What I’m Going to Do About It
The basic concept behind YK is clearly not broken. Prior to the market slide in October, we were able to pull gains from up, down, and sideways markets that I’ve never seen a fund trading program do in real-time relative to its downside volatility. So the basic underlying concept doesn’t need to be changed.
But YK needs a mechanism to recognize when the market is wildly out of sorts. I’m pursuing two avenues at this moment (warning: this part of the discussion will be a little geeky).
- The first options is something very simple such as moving to cash whenever the market moves beyond a volatility envelope (such as Bollinger Bands). The disadvantage of this option is that over the long-term it leaves some money on the table because buying into extreme weakness is (on average) very profitable. The advantage is that it without a doubt takes the strategy to the safety of cash when the stock market is doing something out of historical norms (assuming it doesn’t get caught in a catastrophic down day, a’la October, 1987).
- The second is something a bit more sophisticated where in addition to looking at for example average expected returns for a combination of conditions when making its daily trading decisions, the strategy is also looking at outliers (such as the bottom 2% of returns) to try to gauge the potential for large losses given a combination of conditions. The advantage of this approach is that it tries to capture fat-tail risk before it happens. The disadvantage is that it doesn’t necessarily do such a good job at capturing that fat-tail risk accurately – major losses often occur for reasons completely unrelated to technical factors.
So that’s where we’re at today. I banned myself from trying to cobble together a hasty answer this week while we’re still in the fire. I want to make a decision that we can live with indefinitely. YK and I have crunched some preliminary numbers on each of the solutions above, and we’ll be narrowing down the decision in the coming week (though I’m heavily leaning towards the more conservative solution #1). I’ll be updating this blog with our solution as a diary of sorts.
Happy Trading,
ms
To stay up to date with what’s happening at the MarketSci Blog, we recommend subscribing to our RSS Feed or Email Feed.
Filed under: Evolving Markets, Trading Strategies | 13 Comments
I think this week has found loopholes in almost everyone’s risk management strategy. I am down 15% on the month, which exceeds my historical drawdowns by 2 times. Very disappointing, but if you can make it out alive your system will be much better off for it.
We all want that huge ‘collect’ when things look too good to be true. But that’s exactly the right time to downsize, reduce risk, and still target above average returns.
Everyone wanted to short VIX at 40. Those who did, were blown away. Those who nibbled, got to nibble again at 50, 60, 70 and soon 80? And if they were smart enough to own insurance, they not only didn’t get crushed, they probably earned a profit.
I’m sure you’ll find a good adaptive solution, but going ‘conservative’ sounds right to me.
one thing I’ve used for years is the average absolute value of the gap from the prior close to the open. Beyond a certain percentage I’m out of the market regardless of any other indicators. Though it’s a very “dumb” strategy, it doesn’t even look at direction, it’s kept me from getting slaughtered a number of times over the years.
With the change in the YK strategy, let us be sure that it will not put us in an inverse fund for an extended period if the market goes straight up also.
Thanks for writing about this failure. This is very helpful to a fellow system trader/designer.
I pulled up the timer trac graph before reading this blog entry, and I thought maybe timer trac had made some errors. That is nasty looking. Sorry to see it.
One thoughts…
One part of future adaptations could be variable position sizing. Should the system hit a series of losing trades, position sizing is decreased by X percentage on each losing trade.
RE to just doug: good concept…the basic difference is whether (a) all volatility is bad as in your concept, or (b) is just directional volatility bad as in the volatility band concept. I didn’t mention it in the original post, but I’m pursuing both ideas…more to follow.
ms
RE to Peter: absolutely. One of our central tenets in designing YK was to make sure everything was symmetrical…in other words, that we don’t treat long by one set of rules and shorts by another. Any solution should work across the board.
ms
RE to Woodshedder: good comment re: scaling position sizing with consecutive losses. I had this same thought initially. The problem is that none of the statistics support this approach. Contrarian strategies frequently pick the bottom too early and (with the exception of freakish times like we’ve had in the last week and a half or so) this means that the rubber band has been stretched just that much tighter and is that much more ripe for a rebound. So over a long time horizon scaling down with losing trades is going to negatively impact performance.
I’m trying to build something more geared towards id’ing those times when the market is temporarily not following historical norms and thus no longer able to be analyzed using historical data.
Thanks for the comments and the great blog.
michael
Michael, no, thank you for your blog. It is very helpful to me.
I understand what you mean about reducing position sizing reducing peformance over time, but when you are up like what, 70% since inception, it would really really have to impact performance to bring your system back out of the stratosphere…lol…I mean that as a compliment.
One question, that you may or may not choose to answer. Does this system have long winning and losing streaks?
Woodshedder:
RE long winning/losing streaks: yes and no. In terms of winning days, in both historical tests and realtime we don’t do much better than 50/50, so yes, we have long streaks of losing days. The “power” of the strategy comes in the size of our winning vs losing trades. Has to do with the fact that everyday we are taking some kind of position but we’re scaling the position size based on our confidence in that day’s analysis. So a lot of days we’re in the market with a very small allocation b/c the strategy isn’t very confidence (hence the lower % of winning days).
RE the stratosphere: bulls get fed, bears get fed, pigs get slaughtered eh? You’re probably right =)
Is the position sizing, based on confidence of the day’s analysis, part of what makes the system adaptive? Or is it the direction (long/short) it takes?
RE to Woodshedder: Good question. Answer is both (long vs short and degree of confidence) + other elements as well such as what’s important to consider in determining that confidence. michael