Edit: daily signals for this strategy are available in the State of the Market report.
In my last post I tested two long-term strategies from Mark Boucher’s “The Hedge Fund Edge” that were based on measuring the attractiveness of bond yields in order to trade the S&P 500. In this post, I want to offer an alternative (and more effective) short-term trading strategy that exploits this same idea.
The premise of this strategy is that stocks and bonds are always in competition for investors. As bond yields become more attractive, investors move away from stocks and into bonds, which is of course, bearish for stocks.
The graph below shows the S&P 500 (blue), the following strategy (green), and for comparison’s sake, the inverse of the strategy (red) from 1962:
[logarithmically-scaled]
Strategy Rules: Go long the S&P 500 at today’s close if the 5-day exponential moving average (EMA) of 10-year treasury yields is falling. Close the position and move to money market when the EMA is rising. This test is frictionless, and to compare it accurately to Boucher’s strategies, I’ve assumed money market returns equal to half of the nearest 3-month US Treasury bill.
In a nutshell, the strategy is bullish on stocks when treasury yields are falling relative to very recent history, and neutral/bearish when yields are rising.
The strategy did a good job outperforming the market in all periods with the exception of the late 90’s and mid 2000’s when, one could argue based on the subsequent bear markets that, equities became overinflated relative to treasuries. Drawdowns in all periods were managed well.
For the number-lovers, below are stats for this strategy relative to both the S&P 500 and strategy #1 from my previous post:
CLOSING THOUGHTS
A lot of improvements could be made to this strategy (think 5-day EMA relative to a longer-term EMA, measuring the EMA’s ROC, etc.) but I think I can say that this shorter-term strategy as it looks now is “better” than the longer-term version – and not just because of its historical performance.
In my opinion, short-term strategies (when trading assets that minimize/negate transaction costs, such as leveraged funds), are better than long-term ones because: (a) there are more historical trades to study, reducing the degree of curve-fitting, and increasing the probability the strategy will work in real-time, (b) it’s much easier to determine if a short-term trading strategy has stopped working – reviewing the results of say ten real-time trades would take about 2 months for the short-term strategy, but about 16 months for the long-term version, and (c) I’ve found that long-term market inefficiencies are much more unstable and likely to be consumed by the market.
Regardless of your opinion on the above, I think we can agree that strong treasury yields have been bad medicine for equities, even in very short timeframes.
Happy Trading,
ms
Geek Note: There are two generally accepted ways to calculate an EMA that produce slightly different results. Here I’ve used the ((1/Period)*2) method. If your charting program uses the (2 / (Period + 1)) method, simply reduce my period by one. For example, if I’ve used a 5 period EMA, the alternate EMA would be a 4 period EMA.
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, Treasuries & Interest Rates | 25 Comments
-
Subscribe to RSS Feed
Subscribe via Email
My email: michael @ marketsci.com (remove spaces)
Simple question: how do you define “rising” and “falling”? Rising = value today is greater than the value yesterday? Slope?
I used the slope of the 5 day EMA – go long when the slope is down. What did you get for a win %?
RE to dskills: hello good sir…long time no talk. I defined falling as today is less than yesterday (and vice-versa). I don’t code these type of studies at a position level (rather, at a day level), so I haven’t calc’ed the win % on positions, but 53.8% of days from early 1962 were wins (vs 51.0% for the inverse rules). Hope that helps.
michael
RE to dskills: P.S. I edited the post to include a note at the end about what flavor of EMA I used. For the uninitiated, there are two generally accepted methods of calc’ing an EMA that each produced slightly different results. michael
Does the intuition behind this strategy make sense in times of great crisis like the past few months? I mean, people rushed into treasuries reducing yields. Your system would take that to be a signal that stocks are more attractive, and you would have been buying into the crash (perhaps). That’s not necessarily bad, but have you tried it on this year’s recent data?
RE to anon: good comment. Two reponses: (a) this test is current as of Friday 11/14, and you’ll notice that this approach did significantly better than the inverse over that period, (b) we have to be careful not to confuse long-term themes such as the one you mentioned with short-term inbalances such as what this strategy is trading. We’re looking at a only 5-day EMA and while the broader theme in this market might be “falling UST yields still bad for stocks”, at a day-to-day level that’s not always the case. Just my 2c.
michael
Yes, I see what you are saying in (b). It would be interesting to see how the performance of the strategy varies with the length of the time window. So far you’ve looked at long term and short. Of course, one doesn’t want too many parameters.
On (a), yes it does look to be better than the inverse for sure, and from the bit on the graph maybe even the market.
In general, I think all strategies need to be tested rigorously through crises periods.
Hi Marketsci,
Some comments on your short term strategy. I have backtested it from 05/11/1962 to 11/14/2008 using exactly your model.
Input data is from Yahoo Finance (^TNX for 10Y US rate, ^IRX for 13W US T-ill, ^GSPC for S&P500).
What I have found out is that the return is VERY sensitive to the level of commissions for trades. If commissions exceed 1/3000 of your net worth, your strategy performance will decrease dramatically and turn deeply negative if commissions exceed 1/1000 of your net worth. Moreover the assumption of getting 50% of daily cash return is optimistic with small size retail accounts. If you assume conservately that cash does not earn interests, the strategy annualized performance drops to 6.4% for a 11% volatility, which is frankly disappointing.
Can you elaborate on your backtesting to exchange on assumptions ?
RE to Shivar: these are my favorite kind of comments – shows that folks are paying attention.
You’ll notice that most of my tests are “frictionless” because the overarching theme of this blog is understanding the relationship between things – how does “this” affect “that”. To your point, none of the strategies I talk about on this blog are meant to be applied without first carefully analyzing how that strategy fits with what the reader is trading. Are we using leveraged funds (such as from Rydex) where slippage/commissions are not a factor or a small retail account trading ETFs (where slippage would render this strategy useless). How would intraday orders (stop/limits) affect performance? Etc.
More importantly though, I personally wouldn’t trade any individual strategy discussed on this blog by itself. I think investors have two options in trading mechanical systems. First, much more robust observations do exist and can be discovered with a lot of legwork (as evidenced by my own strategies and their audited-performance). Alternatively, concepts such as the ones we discuss on this blog can be combined for a more multi-dimensional approach to the markets (which is the angle I’m offering on the new State of the Market report).
In short, the important take away from the post is that treasuries and the stock market are clearly and definitely related in very short time frames. The specific way that the trader exploits that relationship is a question only each individual trader can answer for their own specific situation.
michael
What is the standard deviation on each strategy?
RE to Scott: which SD are you looking for? SD of trade returns, daily, monthly, annualized, etc? michael
Do you have the annualized versus the buy and hold sd?
RE to scott: sure. To keep things simple, I discount all Sharpes at 4%, so a little backward math shows the buy&hold annualized SD as 15.6%, and the short-term strategy as 11.2%. Hope that helps. michael
Have you investigated a somewhat similar strategy discussed by Pu Shen in a paper ‘Market Timing Strategies That Worked’. He proposed exiting the ‘market’ (defined as s&p 500) when the difference between the E/P ratio of the S&P 500 and the yield of the 3-month treasury bill reaches the ‘extremely low range’ which he defines as the tenth percentile of the historical range of the spread. In sum, the strategy only made 29 actual trades in 31 years (1970-2000) but $1 invested in 1970 ending at $106.69 vs $46.65 of buying and holding the benchmark (excluding trading costs). I would like to track this strategy, but I’m finding it difficult to access the historical data necessary to define ‘the 10 percentile’.
RE to Scott – I haven’t seen that – have a link you can share?
Can I assume you meant P/E ratio or did you actually mean E/P ratio?
Generally speaking, I don’t test or trade stategies that trade so infrequently. (a) I’m inpatient, and (b) with so few trades to study, I think it’s very hard to draw definite conclusions. Having said that, would love to see the paper and put it to the test.
michael
Hi Michael. He actually uses E/P – Pu Shen works (or worked, not sure) for the Federal Reserve in Kansas City. Various links are below:
http://papers.ssrn.com/sol3/papers.cfm?abstract_id=445920
Or direct from the KC frb: http://www.kc.frb.org/Publicat/reswkpap/PDF/RWP02-01.pdf
RE to Scott: thank you kindly sir – been added to the todo list. more to follow. michael
You’re welcome, I’ve added quite a few of your trading strategies to my blog, scottsinvesments.blogspot.com, I’ll let you know if I have any other ideas for you.
RE to Scott: thank you sir – I’ve added you to my RSS reader. michael
Michael, could you suggest a decent source of 10-year treasury yields data please? I looked at Mark Boucher’s stuff a long time back – he’s been at this a fair while.
Thanks
John
RE to John: Yahoo Finance should be fine for treasury yields. Go to finance.yahoo.com – enter the ticker (^TNX I think?) and then click historical data. michael
Yes I found TNX, I also downloaded from http://research.stlouisfed.org I will see what I end up with. Your intial staregy here looks very promising.
Thanks
John