CSS’s “Perfect Alignment” Theory
This is a test of CSS Analytics’ Perfect Alignment theory.
As the theory goes, the market is in “perfect alignment” when: (a) the S&P 500 (large-caps) is outperforming the Russell 2000 (small-caps) over the long-term, and (b) the Nasdaq (high beta) is outperforming the NYSE (low beta) over the shorter-term. Readers will note that the two parts of the theory are similar to the Generals Lead the Troops and Nasdaq vs S&P 500 strategies we’ve covered previously.
CSS lays out an excellent narrative re: why these inter-market relationships exist, but discussing “fundamentals” causes an involuntarily gag reflex in this geek’s belly, so I’ll just be looking at the numbers.
The chart above shows the four possible states of alignment, along with the annualized return and Sharpe Ratio of the S&P 500 the day following each state from 1989 (insufficient data exists to test prior to that). The market has spent roughly 1/4 of the time in each quadrant.
Along the top of the chart is the first criteria for perfect alignment: the S&P 500 leading (bullish) or lagging (bearish) the Russell 2000 over the previous 252-days (1-year). And on the left is the second: the Nasdaq Composite leading (bullish) or lagging (bearish) the S&P 500 over the previous 42-days (2-months).
Note that I’ve changed CSS’s second criteria a bit: (a) I’ve replaced the NYSE Index with the S&P 500 (a near perfect substitute), and (b) I’ve changed CSS’s 20-day lookback to a 42-day lookback because it’s performed similarly and better matches our previous Nasdaq vs S&P 500 test.
Looking at just these average returns we see that (a) the market has been strongest when it was in perfect alignment, and that (b) the S&P 500 vs the Russell 2000 (AKA Generals Lead the Troops) has been the more important of the two criteria.
[log-scaled, growth of $10,000, click to zoom]
In the graphs above (click to zoom) I’ve broken out S&P 500 returns the day following each state. These results are frictionless (i.e. ignore transaction costs and slippage) and do not include return on cash, but could have been reproduced for all intents and purposes using actively-traded mutual funds (our weapon of choice).
Same basic conclusion: the market has been strong whenever criteria #1 was met, but especially when coupled with criteria #2 (perfect alignment). Criteria #2 alone hasn’t been particularly impressive, which the stats reflected, and the market has consistently performed poorly when neither criterion was met (misalignment).
In Summary
Again, I can’t comment on CSS’s explanation of why these inter-market relationships exist, but they have consistently held true since at least 1989.
Note that criteria #1, the S&P 500 vs Russell 2000 (AKA Generals Lead the Troops), is available daily on the free State of the Market report.
[click for a follow up to this post with a different spin on CSS's theory]
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: Trading Strategies | 4 Comments
Michael,
You have a wonderfully clear and simple way of presenting strategies. I’m always impressed by how clear you make it.
I’m with David on the value of an underlying model though. One can always find an equation to fit a set of historical points. And one can never know whether the equation will have any predictive value. The only way to develop real confidence in a strategy is to understand the dynamics that it is taking advantage of. (Although I’ll admit that if a strategy that is developed through back-testing continues to produce profits, that’s also a pretty good confidence builder.)
There is always Chris Anderson’s claim that data makes theory obsolete (http://www.edge.org/3rd_culture/anderson08/anderson08_index.html). But I’m not buying it. Data helps perception immensely. But it can’t eliminate the need for understanding.
RE to Blue: thanks for the kind words sir.
I agree with you re: the need for fundamental attribution (surprise)…I just don’t feel that my quantitative brain is really qualified to have that discussion. At the end of the day, I’m pretty good at the specific thing I do, but there are a whole mess of folks (yourself included) that understand fundamentals better than I.
michael
I am fixing here the messed up text of my previous message:
Michael, your tables are a good deal counter to what CSS’s results are: the top left-bottom right diagonal is in both of your cases similar (for you strongest: SP>Rut & Nq>SP; weakest: SP<Rut & NqRut & Nq>NYSE and wekest: SP<Rut & NqRut& SP>Nq while CSS’s #2 strongest results come when SP>Rut & Nq<NYSE. Correspondingly CSS’s #3 strongest results come when SPNYSE while your #3 strongest results come when SP>Rut & Nq>SP. Assuming an equivalence between SP~NYSE the difference in results in the two cases of the so-so allignment is impressive. Can it be explained just by the difference in your time frame forthe secondary criterion (you 42d, CSS 20d) alone?!
RE to kostas: I think there might be some confusion b/c my table is in a different order than CSS’s. Take a second look – the two actually match up very well. michael