The Abnormal Market Filter Says “Barricade the Doors”
The Abnormal Market Filter reading for Tuesday (02/24) has moved to 100% for the first time since the market meltdown last October.
I’ve talked a lot over the last few months about my mechanism to determine when the market is acting “abnormally”, the Abnormal Market Filter. The idea behind the filter is to give my own programs a way to measure when the market has moved beyond normal ranges, because strategies based on historical norms might not be suited for that very abnormal moment.
The filter is included in the State of the Market report and both the YK and Scotty strategies, and in all three cases, as the market becomes more “abnormal”, position sizes are reduced (eventually to zero). The graph below shows the S&P 500 (blue) versus the filter reading for that day (red) from the beginning of this year.
During both of this year’s market slides, the filter ratcheted up, hitting a high of 75% in January and 100% at this moment now.
These two very elevated months have given me an opportunity to look at how our programs react in real-time to this idea (and so far, I’m a very satisfied developer).
To illustrate, the graph below shows YK’s performance YTD both with (red) and without (green) the abnormal market filter. The red line represents the strategy as we issue it to investors, and the green line the old pre-filter version of the strategy.
[linearly-scaled]
For most of the year, the Abnormal Market Filter has actually reduced portfolio returns, and I think generally speaking this is going to be the case; there is significant money to be made when the market is making big volatile moves.
But there is also significant pain to be inflicted when we’re wrong, and the filter has done a very good job this month defending against a series of what would have been very bad positions. YK returns without the filter would have been roughly -10.1% month-to-date. With the filter? -0.9%. That’s a substantive difference.
I have no idea where the market goes from here (I leave market calls beyond 24 hours from now to the pundits). By design, the Abnormal Market Filter returns to “normal” very quickly as the market stabilizes so I foresee taking bigger positions again shortly, but at this moment in time, I would be very hesitant to make large bets in this market.
Note: Readers looking to follow the abnormal market filter can find it every day in the second box of the State of the Market report (click to zoom).
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: Trading Strategies | 20 Comments
Just a question on the principle behind the abnormal filter.
Does it depend on an exogeneous variable (for instance S&P500) or does it depend on the performance of your model itself?
In a former post, you elaborate on “bollinger bands” like principle based on volatility of S&P500. Which can be used to adapt the leverage of YK strategy.
Or you could say: ok I expect from a statistical perspective that there is only 5% chance to see such a drawdown in such a few days (before filtering), so I reduce the size of YK according to the probability of occurrence of the drawdown. Which is a bit like studying the autocorrelations of YK daily return.
Shivar
Fantastic work Michael!
Michael, I’ve added an ab. filter to our systems as well. The filter does add a bit of psychological comfort, albeit at the expense of returns. However, as I know full well, over-riding a system will often damage returns a lot more than having an ab. filter. Having such a filter I believe can allow the hard trades to be placed knowing that at some point, the slide into the abyss will be mitigated, or even stopped outright.
For such a simple addition to a mechanical system, I believe your contribution of this idea has had perhaps the most the benefit to my development, followed closely by adaptive systems. Thanks!
RE to Shivar: good question. It’s the former not the latter (based on the market, not the model). As the market moves farther and farther beyond normal recent ranges, the model takes smaller and smaller positions (regardless of how successful it has been or what direction it’s trading).
michael
RE to Wood: glad to hear it sir. I think you hit the nail right on the head with my own results. Historically, my own ab filter has a slight negative impact on returns, but the psychological comfort (and reduced drawdowns) of knowing you’re paring down position sizes when the market is going crazy is well worth it.
P.S. still doesn’t resolve the single day bomber (a’la October, 1987) – this will be a topic of future posts this year as soon as I can get my act together.
michael
A wise trader once quipped to me that “one man’s stop is another’s entry point.” I think he meant to imply that successful stop strategies should be used as system reversal signals, causing you to take the other side of your prior position. Thoughts?
RE to Market Rewind’s Jeff: he was a wise man indeed.
This is the hurdle that made me for so many years averse to adding something like this to my programs. On average, over enough historical data, the farther the market gets stretched, the more likely (and more violent) the reversal should be. So any attempt like the Abnormal Market Filter to pull the strategy out of the market as it becomes more and more stretched is going to negatively impact backtested (and I think real-time over a long enough horizon) performance.
The problem with that very analytical approach is that it fails to capture the psychological pain (and potential catastrophe) of trading the market crash. That could be to the individual investor, or in my case, to the third-parties who use my programs and I am accountable to.
I’ve changed my trading philosophy really since the October crash last year to be: I generate a lot of consistent positive alpha – why am I getting greedy and trying to pull out these extra gains when the market is clearly in distress? Better I think to take my returns when the going is good and step aside when times are troubled (even if it means that over the long-term I have to sacrifice a bit of those returns to do it). I guess that’s the wisdom of old age creeping in =)
michael
Hi Michael,
Thank you for your interesting feed back. Just for the sake of clarity, do you aim to cross validate “abnormality” with using multiple exogenous variables or do you at the moment look at fluctuations in S&P500 and derive your abnormal filter.
I had thought for instance to cross validate abnormality with looking at simultaneous daily ranges in a set of key macro variables: S&P500, 10Y rate, gold price, commodity index price … With the basic principle: if all these assets move out of their normal range together, then there must definitely be strong abnormality.
Regarding your desire to dodge the 1987 krach, I would like to make a few comments:
- as far as I know, no crash comes as a completely unexpected event. There is no real black swan. If you look carefully back to 1987, your abnormality filter would have been more than triggered before the fatal day of the krach (in my backtest: abnormal filter at 90%).
- most of the times big daily moves go in the same direction as the long term trend (but can generate a strong temporary reversal in the opposite direction). So if your strategy is structurally long volatility and long trend, you should dodge or profit from krachs.
Shivar
And there you have it, thanks for closing that loop.
It would be nice to see what the raw system says without the abnormal filter … I understand why you need the abnormal filter, and why the focus should be on the final, adjusted signal, but it is awfully nice to see the various bits and pieces even when the abnormal filter is @ 100% …
RE to David: when you say the “raw” system are you referring to the State of the Market report, and if so, to the aggregate predictions? I was kind of torn over showing the aggregate prediction with and without the abnormal filter, but decided on just showing it post-filter to keep it easier on users.
The individual strategies tracked (the second box in the State of the Market report) is NOT adjusted by the abnormal filter.
Please let me know if I’ve misunderstood you.
michael
RE to Shivar: I’m currently only using price data from the S&P 500 itself, but the concept you’ve presented seems perfectly reasonable as well.
I do agree with you that the abnormal filter would have taken the portfolio out of the market almost entirely in the crash of October 1987 for instance, but I don’t agree that this will always be true in the future. Heaven forbid should an earthquake hit tomorrow and California floats out into the Pacific, the Abnormal Market Filter isn’t going to protect the portfolio. Just my $0.02.
michael
Hi Michael,
I am afraid not to see the point that you are making on extreme outlier events.
If I am not mistaken, the objective of an abnormal filter is to prevent YK to invest when the market BECOMES abnormal -> if the market starts behaving abnormally, then the model should stop bets as soon as possible.
It does not protect you against the occurence of the initial event which triggers abnormality (which cannot be detected as you explain fairly well). But this is not the purpose of an abnormality filter. The solution would be rather in size of bets to insure that in no case the max instantaneous drawdown would be too harsh. And/or it is to structure bets in a way which protects you against the behavior of markets in an extreme case
-> typically equity market is skewed to the downside, commodities (including gold) and bonds to the upside. So YK applied simultaneously to these three assets classes with constraints on the overall allocation could use these different skews to protect you against extreme events.
Buying insurance/out of the money options is not a money making idea if you have no anticipation that such event should happen soon.
Shivar
RE to Shivar: your idea RE: applying YK to these very different asset classes is interesting, and something I would love to do, but not possible for me right now.
Generally speaking (and this is just my personal opinion), bonds, gold, and other non-equity investments are incredibly difficult to time in the very aggressive “everday” fashion that YK uses. I’ve seen bonds done well in short-term trading, but I’ve never been able to figure out how to compete in that arena.
I think the options overlay (at least based on the preliminary numbers I’ve seen out of Jared with Condor Options whose helping me flesh out my thoughts on it) can make sense for a program that generates as much consistent alpha as we do. That’s really the kicker to making the idea fly right, generating enough return to eat the overlay costs.
Great comments…thanks for the thoughts.
michael
PS to my note to Shivar – remember too, that the options overlay isn’t about making money. The overlay is about protecting the portfolio from a massive single day loss (a’la October 1987). michael
Hi michael,
Just a few comments on your answers:
- regarding bonds for instance: bonds show a very strong positive autocorrelation, therefore their returns are far more predictable than equity returns. Any idea ala YK should perform far better on bonds. For commodities, the balance between the “trend following” part, the “mean reversion” part and the “cyclical” part of the forecast is different than the one for equity, and therefore provides you with a lot of diversification in model risk.
- purchase OTM options is an expense. If this accrued expense is not compensated by the income generated during a krach, then it is not worth the effort. That is how Taleb blew out all his hedge funds. Moreover it is an idea very “trendy” today as risk aversion is very high but by definition options are very expensive when risk aversion is very high. I really think that you can make profitable daily trading with options but should not use them as an insurance overlay if you have no specific view on the market (if you have a view, it is all different of course). The insurance protection is to be found in proper diversification across time horizon, asset classes, class of models and a part of own decision making.
RE to Shivar: I won’t debate the comments re: the overlay – not my field of expertise and something I’m leaning on other experts to help me flesh out.
But I will take the bond one. Do you have any data on the autocorrel. comment? That doesn’t jive with what I’ve seen. I’m can’t believe anything would exhibit the level of day-to-day mean reversion that we’re seeing in equities indices right now (ex. http://marketsci.wordpress.com/2009/02/15/short-term-mean-reversion-becoming-stronger-part-iv-so-what/).
michael
Michael,
Have you given any thought to allocating a small part of your portfolio to a VIX type ETF in lieu of the AMF? My thinking is that you would avoid the performance hit from reducing your exposure with your other strategies when AMF kicks in and offset any losses with jumps in in the VIX ETF. CXO featured this research paper recently: http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1663803
Would love to hear your thoughts.
-Beau
Hello Beau – good question, but in short, no.
Volatility has on average been very profitable for us (and a lack of volatility, less profitable), with a few exceptions. It’s those “few exceptions” that I’m concerned with (hence the reason we ratchet down exposure).
Put another way, the return vs volatility equation (in my specific case) isn’t cut-and-dry so I wouldn’t want to play volatility as a straight hedge.
michael