Strategy Mashup: Trend Following + Monthly Seasonality
I’ve received quite a few requests to show the results of trading a combination of two strategies I’ve talked about previously: a simple trend follower (50/200-day moving average crossovers) and monthly seasonality (the turn of the month and options expiration week).
I think many readers honed in on those particular strategies because they’re (a) much less active than much of what we talk about on this blog, and (b) so easy to calculate and trade.
So by reader demand, your strategy results in two different flavors (note: all of these strategies are tracked daily as part of the free State of the Market report):
[logarithmically-scaled]
The graph above shows two flavors of the mashup strategy (green and red, more on this later) versus the S&P 500 (blue) from 1988. I’ve assumed both the trend-follower and monthly seasonality strategies were traded long-only, and because I want to reward the strategies for a lot of time spent out of the market, I’ve assumed a return on cash equal to half the nearest 13-week Treasury.
The results are frictionless, but for all intents and purposes, could be duplicated today using actively-traded mutual funds (the only thing I trade, not to be confused with ETFs) minus an annual fund expense ratio.
And for the number lovers:
Flavor one of the strategy assumes that a long position is only taken when both the trend-follower and monthly seasonality agree. Flavor two splits the difference and assumes we trade half of each day’s position according to the trend-follower and half according to monthly seasonality.
Both flavors have significantly outperformed the market over the last 20+ years. I’m especially pleased that they weathered both major downturns in the market (2000-2002 and our current one) so well.
Two things that I don’t like about trading this mashup:
First, as readers know, my programs are all very active swing trading strategies; psychologically, I couldn’t deal with being out of the market for such long swaths of time (but that’s just me).
And second, 50/200-day crossovers have a very, very long history of risk-adjusted outperformance, but as I talked about when I first introduced them, the monthly seasonality strategies didn’t perform well prior to 1988. Normally, with a very short-term observation like adaptive daily follow-through, that wouldn’t bother me, but for this type of non-price-related observation it does (point being trade with care).
Happy Trading,
ms
P.S. Note that all of the strategies discussed in this post are tracked daily as part of the free State of the Market report.
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: Time-based, Trading Strategies | 16 Comments
I’d like to see the spreadsheet calculations of these models. I’ve tested all of this stuff and I’ll bet anything they’re suffering from look-ahead bias (i.e., calculating the strategy return for day x using the closing value on day x, when in reality you can’t base your decision to trade at the beginning of a day based on that day’s closing value).
RE to ajk: I am 100% sure that the test does NOT suffer from look ahead bias (and a little annoyed by the snarky response). I will send you an email with a simple excel worksheet in the next couple of days when I get a little time. michael
Thanks for the reply. I didn’t mean to come off like that (though re-reading my response that’s the only way it could’ve come off). Clearly you are the expert and not me. If everyone could be successful at this then it would cease to exist as a profitable opportunity! I would appreciate the follow-up for my own edification.
Sharpe Ratio should be unitless.
RE to shocklee: (stating the obvious I know, but…) 9.8% is the same as 0.098. Anyone who can’t figure that out couldn’t understand or apply the concepts from this post anyways. Besides, I like my columns to line up nicely =)
1. So if you have a Sharpe Ratio of 9.8%, what is it a percentage of? 9.8 percent of what?
2. By your logic, if the S&P 500 index is trading at 735.09, we can also express it as 73509%, right?
RE to shocklee: nope, the difference is the Sharpe Ratio is a RATIO (in this case of annualized return less a discount rate over annualized standard deviation). The S&P 500 is NOT a ratio.
Here’s a wikipedia entry that might help you out:
http://en.wikipedia.org/wiki/Sharpe_ratio
michael
Okay, here’s a different example. Ever heard anyone say the P/E ratio of a stock is 2300%? Nope.
Looking at the Sharpe Ratio wikipedia article, it reflects my exact point that SR is a unitless number not expressed in percentage terms.
How different would you say your trend follower strategy be from the 200-day moving average strategy outlined by Jack Ablin of BMO and described in Ritholtz’s Big Picture blog on November 14, 2008 entitled, The Lost Decade? (http://www.ritholtz.com/blog/2008/11/the-lost-decade/)
Have any idea of which might have produced a higher return over your 1988-2008 time period?
RE to stockchartist: funny you should ask – covered that same question (based on that same post) here:
http://marketsci.wordpress.com/2008/11/18/test-of-ablin-trend-following-strategy/
michael
Hi Michael,
Just for information, you can improve significantly the monthly mash-up strategy with defining rules of investment which depend on the month. Typically March or December are good months, September is awful. I have improved it from ~ 10%/year return to ~18%/year return.
Shivar
I have just discovered your site and really enjoy it. I am curious, for your chart and returns table above, what criteria are your using to exit the trades?
Thanks. W
RE to Shivar: good thought. I did a couple of reports on seasonality through the year back on the old MarketSci site, but I don’t think I’ve done anything on the blog (or at least I couldn’t find it). For longer-term strategies like 50/200 crossovers, I think that type of seasonality might make some sense. For very short-term strategies like the proprietary ones we offer I don’t think it does because we’re playing advantages that are in such a different time scale.
RE to Wink: when the entry criteria is no longer valid. You’ll have to click through to each individual strategy (50/200 crossovers or monthly seasonality) to see the specific rules tested.
michael
This analysis is worthless unless you prevent curve-fitting by using random-sampled data, out-of-sample data, and monte carlo analysis. I have 1,000 trading systems that work on historical data, but don’t make money. You have to do the full analysis.
I’m actually surprised that I don’t get this comment more often. I break from the conventional wisdom on this issue. Allow me to explain…
For very simple single (or near single) rule systems like this one, I’m concerned with the consistency of returns over the entire sample, not in/out of sample testing. Think about this for a moment – if you cut that test in half and only looked at the first half, would you conclude it’s a good system (at least relative to the whole)? Of course. So there’s your in/out of sample test. You’ve accomplished the exact same thing…means absolutely nothing.
I do not think that’s true however for highly optimized systems. To combat curve-fitting, I absolutely agree that an optimization over in-sample and then walk-forward on out of sample is required.
As to Monte Carlo…hmmm…mixed feelings here. The problem I’ve always had with MC is that it understates the likelihood of catastrophic market slides like we saw late last year and early this one. In an MC analysis those will appear to be statistically unlikely events. In actuality, they are inevitable norms.
This isn’t an attempt to convince you that my way is the best – I design the way I design. I always tell folks that if my break from conventional wisdom is too appalling to wrap you head around, take a look at my real-time independently-audited track record. If that doesn’t convince you I know what I’m doing then we can just agree to disagree.
Thanks for the comment,
michael