Notes and Revisions re: TAA Backtest
Apologies for the delay in this series. The Other had a little unexpected time off and we took a quick jaunt to Macau. Neat city, but doesn’t compare to my second home Taipei (台灣加油). Back to the geekery.
My last post showing the backtested results of my take on a TAA model received a GINORMOUS amount of traffic, no doubt most of it salivating over the performance of such an easily executable strategy…
…and that always worries me.
A Word of Sanity
As readers know, I don’t show the backtested results of our proprietary strategies (i.e. strategies we offer via managed account or subscription) because it goes against everything I think investors should consider when choosing an investment.
Despite best efforts to keep tests fair and bias-free, all backtests (and by “all” I mean ALL backtests) are subject to curve-fitting (read more).
As an investor myself, I wouldn’t even begin to consider trading a strategy in the real world unless I could tolerate half the backtested return and double the max drawdown (and that’s assuming I completely trusted the veracity of the tester).
My point is turn DOWN excitement over backtests and turn UP excitement about the concepts behind the backtests and then the real-time actual results that follow, because at the end of the day, all that matters is the proof in the pudding.
China and 20/20 Hindsight
Savvy readers noted that it wasn’t really fair to include China Equities in the test because it wasn’t until very recently that China could have been considered a “major” asset class. Good point.
In my test, China Equities traded in only 8% of months with an average allocation of 14%. Taking China out of the mix had no significant impact on results, so the China addition was inconsequential.
Having said that, what would have been a better test is only trading say the top 3 economies from the previous year at that moment in time. Testing using this approach (which would include Germany until EO 2008 and China thereafter) actually improved results slightly by providing more high-beta equity-related data points to choose from. Again, non-issue.
On a tangentially-related subject, I dropped oil from the mix of asset classes. The strategy “worked better” with oil, but it was hard to justify in my head including oil as a “store of wealth” in the same vein as say equities or gold.
Revised Backtest
Unrelated to the above discussion, I’ve continued tinkering with the strategy and thought I’d share revised results before we get to a more technical discussion.
Yes, I know these results are a bit “worse” than the previous, but I think it will make for a better strategy in real-time.
[logarithmically-scaled, growth of $10,000, monthly-interval]
[based on month-end values]
What’s Next?
In my following two posts I’ll take a more technical look at what I considered when building the TAA model. As always, more to follow.
[Edit: click for a summary of all posts in this series on TAA]
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: Tactical Asset Allocation | 16 Comments
Looking forward to the crunchy stuff Michael!
Curious as to how you are running the backtest? Excel spreadsheet? If so, would you be willing to post or provide the template? Looking forward to learning more about how you are implementing the strategy. Particularly interested in how you look at the risk of a trading range market where you get whipsawed back and forth…
Hello Mike – unfortunately won’t be sharing specific code, but will be sharing logic behind the code shortly.
I think this model (like any rooted in trend-following) is going to sometimes get whipsawed in specific assets traded. But I think the advantage of TAA is that you have so many diverse assets that are at different points in the trend “cycle” that when one asset is range bound the damage is mitigated by another asset that isn’t (assuming that even in crises there are enough assets classes being traded that aren’t overly correlated to the range bound one). michael
Michael — I am intrigued and impressed by what you are doing. One question — you probably considered this already, but would a global ETF (like VEU) and a currency (US dollar, Euro, Swiss Franc — take your pick) work in your model, or would there be too much correlation.
Hello Jeff – thanks for the kind words. VEU is definitely too highly correlated (which is why I’m looking at individual large economies).
I’m still mulling USD. I’ve tested it and it didn’t really help or hinder results, but that’s not what has me hung up. Unlike the other assets traded (with the exception of perhaps the commodity index), it’s not a “store of wealth”, it’s a hedge, and as such, it’s not implicitly expected to appreciate over time. So I don’t know if it makes sense to consider in a TAA model. Still thinking on this one.
michael
Michael, could you write down the specific list of ETF’s that are involved here, or is it too much to ask? We already know 4: SPY, GLD, VNQ and IEF, and if you exclude China & Oil, we have only 2 left: Japan & Commodities. EWJ and GSG, perhaps?
Anyway, nice take for the TAA, just as always, impressing and a bit doubting :)
Cheers, Balazs
RE to Balazs: SPY, GLD, IEF, VNQ, FXI, EWJ, EWG, DBC
michael
Hi Michael
What happens to the equity curve comparison when you start in 1975? There is a kicker of outperformance at the beginning of your sample, which affects the rest of the comparison via compounding.
Do you go short for every asset class also?
Oil is a store of wealth, I would say more than gold, as it has a value in use, rather than just looking pretty on girl’d necks /fingers. Imagine you had an extractable oil reservoir in your backyard. It would be a store of wealth.
Thanks
Rod
Hello Rod – just eyeballing the chart, as you would expect in a bullish period, the strategy and the market would match each other pretty well in terms of pure return (until the 2000 bear)…but as I talked about in earlier posts, that’s not really the point of the strategy. Drawdowns and volatility would have been drastically reduced over straight buy & hold.
No shorting.
While I agree with you re: the true value of gold (don’t let the Other hear that), my belief about the sillyness of attributing value to the yellow element is arbitrary. Gold is clearly a store of wealth. Oil less so.
michael
current mix makes more “sense” than last. only thing is I’d substitute either EEM or VWE for FXI.
Hello Steve – such broad indices (even though they may be geographically distant) have consistently been too well correlated with the US market. I like the idea of targetting specific markets much better. The problem there of course is the volatility that comes with smaller and smaller markets (hence the reason why I’m sticking with just the big 3). michael
Be aware of the tax ramifications for ETFs that contain Futures Contracts. An ETN such as DJCI or UCI might be a better choice. Again, I mean strictly from a tax perspective.
RE to Ernest: because of the nature of the strategy, I would expect very few trades to quality for LT CG rates, so I don’t think in this particular case ETFs vs ETNs is an issue. I am working on a variation on the strategy that attempts to hold trades for at least a year, and in that case, yes, it would be a consideration. Thanks for the comment, michael
Michael,
Re: Half the performance/double the volatility
Truer words were never spoken. As a money manager myself, always under-promise and over-deliver (hopefully)!
Just wondering – how would your strategy performed in any random 5-year period within the 1970-2010. Any statistics on this?
Michael,
I really appreciate your model and your work. Thanks!
What I’m most interested in the moment is the topic of selection of asset classes for these TAA models. I have wavered from using all 80 asset classes and largely ignoring correlation (as CopStrat.com does), to your side of things where correlation is key. Could you talk about this topic, how you choose asset classes to mathematically minimize correlation, and why adding more ‘differently’ correlated assets (why not 15 or 50?) may not be as good?
Hope you understand the jist of my line of questioning here…
Thanks