Can adding some complexity squeeze out some more performance?
In my paper, I examined how a simple timing method on 5 asset classes could improve the returns vs. a passive buy and hold allocation. I want to explore an additional area of research – timing the components of each asset class.
The 5 asset classes used in my study were the Standard and Poor’s 500 Index (S&P 500), Morgan Stanley Capital International Developed Markets Index (MSCI EAFE), Goldman Sachs Commodity Index (GSCI), National Association of Real Estate Investment Trusts Index (NAREIT), and United States Government 10-Year Treasury Bonds.
Instead of timing the entire asset class, it is possible to dice each asset class into smaller segments. Below is an example of the timing model on just the MSCI EAFE Index:
Instead of just timing the MSCI EAFE Index, why not apply to model to all of the constituents of that index? Japan, UK, France, Switzerland, and Germany make up roughly ~80% of the index. Below are the results of the timing model on the constituent countries:
Now, what do the results look like at the portfolio level? I equal-weighted each of the constituents, and the results are below for the passive equal-weighted buy and hold (B&H), and the timing applied to the constituents (Timing).
As you can see, the equal weighted portfolio approximates the MSCI EAFE Index fairly closely. Both of the timing models have similar return figures. Most important is that the timing model applied to the constituents is superior to the timing model on the index for the risk measures of volatility and drawdown, resulting in a higher Sharpe ratio.
This study could be repeated for the constituents of each asset class.
ETFs corresponding to the countries discussed in this column are:
Japan, EWJ
UK, EWU
France, EWQ
Switzerland, EWL
Germany, EWG
It seems the graphes don’t show properly.
ditto – no images visible
Apologies, about every fifth post Blogger deletes all of our pics after a few hours. I have no idea why it does this. . .
Suppose that you work 25 positions in the same way. Won’t the number of trades go up too? Maybe the increased cost of trading eats up the improvement?
Something to consider is the increase in transaction costs. Mebane, in your testing do you normally account for commissions? Keep up the great work!
In looking at your timing models, it appears that, as long as there is some diversification benefit, it will always be better to have more positions rather than fewer simply because a greater percentage of one’s assets are more likely to be invested at any given time.
Given that, could you (or have you) considered short exposure to the same positions using the same methodology? For example, lets say I used SH (short SP500) as essentially it’s own asset class – going ‘long’ the inverse using the same 10 month sma triggers. Is there a reason this wouldn’t work?
More positions will increase transaction costs. However, the system is fairly inactive, and results in less than one round-trip trade per asset class per year.
Most investors rebalance their portfolio periodically, so this amount of trading should not affect results that much- but obvsiously depends on the investors account size and trading costs.
Many mutual funds at brokerages such as Fido have 0 trading cost funds.
I don’t really advocate shorting. You get compensated for owning risk premia, that what capital markets are formed on. You have to beat a bogey (bond yield, which currently is around 5%), and do so with less risk. . .Add in the fact you may or may not receive short rebates, etc. But, I have not spent much time testing the theory either.
Interesting elaboration on your approach, but if you’re going to go in the direction of a little more complexity why not use tactical asset allocation with a large universe of assets, namely the constituent parts. And, I will postulate that the market environments, defined by depth of corection in total mkt averages, can be matched to different roc periods for optimal return. Gosh, there are so many parameters to play with. Are you putting any of your models, or featurs of your models to work as a money mgr?
I’m confused. What is the Timing 2X column? The Std Dev skyrocketed and the Sharpe dropped dramatically.
Anon – Yes I follow similar models in our RIA.
Tomk – 2x represents the leveraged model (2:1 leverage including the broker call rate)