This weekend I was pretty busy relaxing! But all of that relaxing did lead me to consider one idea, that it might be possible to capture "participation" as an index filter using "small universe" indexes vs "big universe" indexes. Basically, breadth in its crudest form.
I also wanted to try this with Australian data.
Most on this forum probably know of the STW ETF which tracks the ASX S&P 200 index. But, at least personally I wasn't aware that on the same day, SSGA also released another ETF, SFY, which tracks the ASX S&P 50 index.
So in this case, STW will represnt our "big universe" of stocks, to measure the participation across a basket of equities while SFY will represent the "small universe" and also act as our trading vehicle. One thing I like about large caps is that their largeness allows them to withstand the business cycle better, whereas a lot of small caps will take it in the neck on factors out of their control during a downturn.
I do realise that using a cap weighted index for the measure of participation is not optimal. But for a quick Sunday night quantification, it'll do!
Any good experiment should have some controls. So:
SFY buy+hold returns 2002-08-27 till current
STW buy+hold returns 2002-08-27 till current
Annualized Return 0.0562
Annualized Std Dev 0.1770
Annualized Sharpe (Rf=0%) 0.3174
SFY using SFY 120day Rate of Change > 0 as index filter:
Annualized Return 0.0509
Annualized Std Dev 0.1733
Annualized Sharpe (Rf=0%) 0.2939
SFY using SFY 180day Rate of Change > 0 as index filter:
Annualized Return 0.0491
Annualized Std Dev 0.1203
Annualized Sharpe (Rf=0%) 0.4086
SFY using SFY 240day Rate of Change > 0 as index filter:
Annualized Return 0.0608
Annualized Std Dev 0.1193
Annualized Sharpe (Rf=0%) 0.5101
I'm providing a range of lookback periods, mostly just to demonstrate that the lookback period isn't too important (i.e. you can derive the same results by any method which captures momentum/trend).
Annualized Return 0.0640
Annualized Std Dev 0.1189
Annualized Sharpe (Rf=0%) 0.5381
As a general rule of thumb, longer lookbacks will underperform shorter ones during bull markets and vice versa in bear markets.
Great. Nothing new to see here! Now. What happens if we add more information, about more stocks into the above crude trading system? What I mean is, what if we use STWs Rate of Change as the trend filter for SFY?
SFY using STW 120day Rate of Change > 0 as index filter:
SFY using STW 180day Rate of Change > 0 as index filter:
Annualized Return 0.0767
Annualized Std Dev 0.1217
Annualized Sharpe (Rf=0%) 0.6299
SFY using STW 240day Rate of Change > 0 as index filter:
Annualized Return 0.0856
Annualized Std Dev 0.1211
Annualized Sharpe (Rf=0%) 0.7065
Now, like a good scientist we will also need to test to see what happens when we using SFY as the signal for STW, to make sure we aren't crazy:
Annualized Return 0.0577
Annualized Std Dev 0.1200
Annualized Sharpe (Rf=0%) 0.4807
STW using SFY 120day Rate of Change > 0 as index filter:
STW using SFY 180day Rate of Change > 0 as index filter:
Annualized Return 0.1206
Annualized Std Dev 0.1090
Annualized Sharpe (Rf=0%) 1.1072
STW using SFY 240day Rate of Change > 0 as index filter:
Annualized Return 0.1147
Annualized Std Dev 0.1097
Annualized Sharpe (Rf=0%) 1.0456
Wow! That was unexpected! From the initial experiment, it looked like using STW as a signal for SFY was a good idea, with better Sharpe than using STW as it's own signal. However there was a little warning sign here, as the 240 day lookback did not outperform.
Annualized Return 0.0884
Annualized Std Dev 0.1109
Annualized Sharpe (Rf=0%) 0.7969
But then when we performed our second experiment we found that, actually, SFY is much more informative to STW than the other way around!
With our scientist hat on, our only role is to come up with some hypothesis and then test reality to see if it matches our hypothesis. If not, we can discard it and come up with a fresh one for testing.
In this case, I would hypothesise that one of two things may be occuring:
1. It is probable that the cap weighted STW is not a good measurement of participation since the top 10 stocks make up >50% of the index value. I hypothesise that a equal weight index would provide a more robust measurement of participation. Unfortunately there is no existing ETF with enough data for me to test and I don't feel like generating my own index on a Sunday night.
2. It is less likely but still probable that due to the nature of the Australian economy, the companies in the ASX S&P 50 actually are more informative about the future performance of the remainder of the market and hence outweigh any "breadth" effect.
In either case, we can derive one small conclusion: namely that using the returns of a related index seems to provide better risk adjusted returns than using the index itself.
Now this is probably as far as I'll take this idea, at least until I get the chance for more relaxation! But since I know everyone loves pictures, here are a couple:
Equity curve of most performant system tested (STW using SFY 120 ROC as signal):
Sharpes as graph: