Trading Strategy: Monthly Seasonality Mash Up
In this report I’ll combine the two monthly seasonality observations I talked about recently, options expiration week and the turn of the month into a fairly effective strategy using nothing other than the time of the month.
First, a review. Historically, the beginning and end of the month, as well as the week (mid-month) leading up to options expiration have been bullish, and the time in between, bearish. Conceptually, the average month might be thought of like a W.
The following graph shows the results of a portfolio trading the S&P 500 long-only (red) using the combined rules from the two previous reports – long the first three and final four days of the month and the week leading up to options expiration – versus buy and hold (blue) from 1988.
[logarithmically-scaled]
And for the number lovers:
Over the last 20+ years, the combined monthly seasonality strategy has outperformed the market in terms of absolute and risk-adjusted returns and significantly reduced downside volatility, while only exposed to the market a little over half the time.
This test was frictionless (no transaction costs or slippage) and ignored return on cash, but these results could be very easily duplicated using actively-traded mutual funds such as those from Rydex or ProFunds (the only thing I trade).
(I probably sound like a broken record, but…) I’ve never been a fan of seasonality plays like this. I think that there are some much more basic characteristics of the market (such as short-term mean reversion) that are much more powerful. But these recent reports have tempered that belief just a bit. I still wouldn’t let monthly seasonality drive my trading, but I will use it as a gauge of the general sentiment of the market.
Both the options expiration week and turn of the month strategies have been added to the State of the Market report as intermediate indicators.
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: Time-based, Trading Strategies | 14 Comments
A simple question … if the state of the market report gives by example -18% What does that mean?
Michael – This is another interesting post for me, in that I am very interested in combining multiple systems to create better, more well balanced trading systems.
I have done a lot of work in this area recently and found the general conclusion to be that that multiple systems will increase consistency & generally product a higher sharpe ratio (assuming the systems are not positively correlated). The return may not be higher using multiple systems, but the variability of those returns almost always is.
Some of the challenges I am running into with multiple systems is how to determine the amount of money to allocate to each one to produce optimum returns.
I think this concept is one people would be very interested in as it is not often talked about in a public arena.
Regards,
Eric
RE to Mr. X: it’s an attempt to assign a “confidence” to the next-day prediction. I can’t give the exact formula (proprietary), but the general idea is that the higher the number, the more confident that strategy is that the following day will be up (if positive) or down (if negative). This number will change over time (that’s the adaptive part of the report).
I use this approach because strategy effectiveness changes over time and at any one time some strategies may be more predictive than others.
michael
RE to Eric D: great topic and I think you could write a book on it. I know that I constantly struggle with this. I think the wrong answer is to do a brute force optimization and see what has produced the best historical results. That approach is just adding curve-fitting on top of curve-fitting.
I think (as you mentioned) correlation has to be considered, how often strategies overlap (or if like this particular post, they’re always segregated), the projected future effectiveness of each strategy…we could keep adding to the list.
I don’t think there’s a magic answer. In the case of MarketSci, I take a static approach based more on logic that optimization. With YK, I take a dynamic approach based on this whole adaptive concept I’m always talking about. And with the State of the Market report I take a dynamic approach based on logic (sort of a hybrid of the two).
I would love to do more posting on this, but I think that outside of a small group of people such as yourself, the significance would get lost on most readers.
Great topic though…
michael
Michael – I agree that there is not best solution for all. I enjoy discussing the topic as it is very complex & there is much to be learned from other peoples opinions.
Over the holidays I am going to read up on some portfolio allocation research to see if the concepts can be adapted to multiple systems. I suspect some of the ideas can but not in a straightforward manner.
I would love to continue the dialog on this offline if you’re interested.
Regards,
Eric
RE to Eric D: absolutely – I’d love to hear any thoughts you might have. One comment re: portfolio allocation. I think that most of the “traditional” approaches are not really applicable to what we do because they assume you have this set of assets that are each always invested. So some of those concepts (such as the impact of correlation on portfolio volatility) are applicable, but I think we really have to take it a step further. Just my $0.02.
All the best,
michael
It looks like the strategy performs similarly to the overall market most of the time, and only outperformed during 1990-1991, 2000-2002, and 2007-current.
Visually, it seems like the outperformance is greatest during bear markets.
I bet the system performs even better when the overall market is below its 200-day moving average.
- Sean
Michael – Yeah that is one of the problems I am running into. There is an interesting webinar on the Matlab website regarding this: http://www.mathworks.com/company/events/webinars/wbnr30357.html?id=30357&p1=56449&p2=56450
They advocate using a genetic algorithm to reduce CPU time to select the best set of equities, systems or whatever to reach some goal for your trading system. I am working on integrating this concept into a system I am working on that selects k trades each day from a population of n choices.
I suspect that the concept of multiple system building will end up going something like ‘given n potential systems to use, if I add system # 1, does it get me closer to my goal. If yes, add it, otherwise do not use the system. Repeat for each system & combination of systems’.
I’ll let you know if I make any notable progress in this area in the upcoming period.
Regards,
Eric
I agree with Eric D, multiple uncorrelated systems tend to display better characteristics when taken together.
I also would be very interested in a discussion on proper funds allocation to the different systems.
Perhaps based on past performance ? But how to measure on an apples to apples basis ?
Excellent postings, perhaps the most rigurous place in the net for trading.
Thanks !
eb
RE to sean: you’re right that it’s definitely not a magic bullet. Remember, the chart is logarithmic, so you have to squint just a bit to understand performance differences.
In terms of risk-adjusted returns, the only extended period that it didn’t outpace the market was was in the mid to late 1990′s.
In terms of absolute performance, just to keep pace with the market, much less more than double returns, with only about half exposure, is a feat. Add in cash returns and I think it’s a workable system.
To your point, not an end-all-be-all, but workable.
michael
RE to eber: WOW…coming from someone who I know (based on your comments) is a student of the markets, that means A LOT.
Here’s the punchline though, I disagree with you.
There are a lot of folks a lot more academically rigorous than I am. I could go that route, but (a) I don’t buy into most of the academic hoodoo, and (b) I think most readers would actually get less value from the posts.
Rigorous? No.
We recently made Condor Option’s Best of 2008 list and I like the way they described us…good at finding “novel” approaches. I like that…I want to slice an dice this market from as many angles as possible (as long as they’ve worked consistently over a long period of time) to make more robust choices.
I guess all of that was a long-winded way of saying “ahhh…shucks” =)
michael
RE to eber/eric: one more wrinkle in this discussion (about allocating between multiple strategies) that I’ve been thinking a lot about lately. As October was a prime example of, historical performance (backtested or realtime) captures “measured” correlation, but fails to capture “potential” correlation. I think any allocation scheme has to include not just what has happened, but what could happen.
So all things being equal, two strategies that always take opposing positions in partially correlated assets (say, a pairs trading strategy) would be preferred over say two strategies that sometimes take similar direction positions, even if the measured correlation between them doesn’t reflect that.
Just ramblings…hope that made sense.
michael
Fantastic post.
On multiple systems, I have devloped a set of systems each one covering different types of market (i.e. trending UP, trending Down and range bounding) this way, at least one system would pick up the market opportunities at any one time.
The adaptive system approach sounds good but i found it hard to “detect” changes in the underlying state of the market though.
On the seasonaluty front, when replacing data with say the NASDAQ or Dow or individual stock, the strategy seems to fall apart, you have any logical explaination for that?