This is a follow up to my post When Adaptive Systems Fail.
In that post I discussed how our adaptive YK Strategy has performed very well in real-time in all different market types – up, down, and sideways – but fell down (badly) when market mechanics made an extreme divergence from history.
The failure was not in assuming that markets don’t follow (evolving) historical patterns – all technical strategies make this assumption and generally speaking, markets do follow history’s (ever changing) footsteps. The failure was in assuming that divergences with history would be brief enough and painless enough that the strategy could ride them out.
So the fix isn’t to try to predict when the market is going to make a sharp, prolonged up/down move (personally speaking, I don’t even think that is possible). The fix is to measure when the market has moved far beyond normal ranges and to then accept that the strategy (which is based on historical norms) might not be suited for the market at that very abnormal moment.
YK and I have approached this problem from a number of different angles – from the very simple to the very elegant – and have settled on the very simple.
When the market moves beyond an extreme volatility band (similar to a Bollinger Band), either high or low, the strategy will move to cash.
Note that this is not the same as moving to cash in times of extreme volatility in the traditional sense of the word. Volatility is not directional – we could experience a huge upsurge in volatility over the course of a month (huge up days followed by huge down days) but still end up net unchanged. YK excels in this environment as we demonstrated in September.
By using volatility bands we are measuring constant volatility in a single direction (unidirectional volatility). We are attempting to capture complete market meltdowns (or “meltups” which very rarely occur) because, by their very nature, they involve markets that are very prone to doing very abnormal things.
I’m unhappy with this solution because (a) over the very long-term it leaves money on the table – trading against extremes is (on average) very profitable, and (b) it adds a partially non-adaptive component to YK – like a Bollinger Band, ours will widen or narrow with the changing volatility of the market, but adding a non-“learning” filter is still very much not adaptive.
I’m happy with this solution because it implicitly and definitely keeps the strategy out of a prolonged market crash. Of course, like any directional strategy, we still bear the risk of a single-day crash (a’la 1987), but that is the implicit risk of any strategy that takes non-hedged positions in the market (and a risk for which we are usually handsomely rewarded).
Last but not least, though it’s all in the past and of no value now (but I know I will be asked), here’s how the strategy would have performed since our real-time track record began assuming perfect execution with (red) and without (blue) the filter, compared to the S&P 500 (green). Note how the filter didn’t prevent a loss (it’s not a crystal ball), but did protect the portfolio when the market went wildly awry:
The question now is when to go live with the revision. Strictly speaking, it should be in effect now. However, this market is so completely beaten up, so prime for a technical bounce, that I have a hard time justifying putting it in place until the next cycle (i.e. it clears and then retriggers).
That’s it…my own brain is in meltdown mode after all of this number-crunching.
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 | 7 Comments
-
Subscribe to RSS Feed
Subscribe via Email
My email: michael @ marketsci.com (remove spaces)
Leaving money on the table isn’t so bad. Your primary goal, as you well know, is long-term success. To achieve that, you must avoid huge hits. Yes, it’s ‘fun’ to prosper during market extremes, but if you want to play, cut size by 75% and play small.
Leaving that money on the table is simply an insurance policy that substantially increases the probability that you will meet your goals. That’s a good thing.
Best regards,
Mark
Possibly a linear regression on the VIX?
I ran this quickly after reading your post and in hindsight, it would have taken you to cash on October 3′rd, which would have saved pretty much the same that you seem to have arrived at.
jog on
duc
Aren’t you trying to statistically model a “Black Swan” event ? That would be a contradiction in terms.
From a more practical point of view, this is a very sophisticated stop loss approach. My impresion is that, like any other variety of stop loss system, it will make you trade drawdown for profits (both will go down).
When living under the memory of a recent catastrophic event, like in 1974, our gut feelings about stop loss systems are invariably warm. Not so much during the halcyon days.
RE to eber: I wouldn’t posture it that way. Am I try to model an extremely sharp 1-2 day event (a’la 1987)? Absolutely not. I agree, it’s impossible. I just want to capture the next time that the market has the potential to slide into the abyss (2008) or into the skies (1999). michael
RE to duc: I think there could have been a number of solutions – that being one of them. I was looking for something that would trigger as little as possible so it didn’t impact the long-term effectiveness of the strategy, but would definitely capture severe unidirectional stress in the market, all while scaling to changing market volatility. We took a lot of sophisticated approaches that I think might actually be better…unfortunately, it’s hard to gain confidence in something that (by its nature) happens so infrequently. In the end, I couldn’t think of anything more straight-forward than a vol band. Thanks for the comment duc.
Jog on.
Michael, what software is creating those charts? They are very well done.
RE to Woodshedder: thanks, but i can’t take any credit. They’re just the standard templates from the newest Microsoft Excel (which by the way, is a bloated, painfully slow hunk of junk…but um, makes nice charts). michael