Performance of the Monthly Seasonality Map
It’s been about 18 months since I first began issuing the Monthly Seasonality Map.
The Map poses a particularly difficult analytical challenge: predicting strong/weak days for the US market, up to a month in advance, based on historical seasonality patterns.
Readers know the Map isn’t a “strategy” per se and seasonality alone never justifies a trade, but I maintain that seasonality is worthy of consideration. To test that assertion, in this post I’ll look at the real-time (out-of-sample) performance of the Map to date.
[linearly-scaled, growth of $1]
The line above shows the hypothetical results of trading the Map since inception in April, 2010, assuming we went long/short (from the previous close) the S&P 500 the same % shown on the Map each day.
The % gained isn’t so important here, returns are simply a product of exposure. What’s important is the consistency of performance.
Numbers for the number-lovers (note: the last column, “volatility-adjusted daily return”, is average daily return divided by the SD of daily returns)…
The bad news: the Map has been correct a bit less than 50% of the time.
The good news: because wins have been so much larger than losses (1.34x), the Map has trounced buy & hold on the stat that really matters, volatility-adjusted return (+7.6% vs 0.9%).
Why? The Map has done well on “strong” prediction days (i.e. predictions closer to +/- 100%), but poorly on “weak” prediction days (i.e. closer to +/- 25%). Hence the reason the win% is so low, but the W/L ratio (and return) so high.
The lesson learned here is that I need to keep doing what I’m doing on strong predictions (ex. first day of the month, Fed meetings, etc.), but I need to reanalyze weaker ones (ex. the Monthly W) to better understand if they’ve just been in a funk or if they’re really not predictive out-of-sample.
Large vs Small-Cap Seasonality
In January of this year, I also began issuing a large vs small-cap seasonality map.
Below I’ve run another hypothetical portfolio, this time pairs trading large vs small caps (S&P 500/Russell 2000) the same % shown on the Map each day.
Geek note: these biases exist after adjusting for differences in volatility between large and small caps (AKA market or beta neutral) so they’re only relevant to either (a) pairs trading, or (b) choosing between large and small caps after an investor already has a view on the market (read more). To stay beta-neutral, I’ve adjusted the allocation to balance the standard deviation of daily returns (over the last 42 days) of each leg in the pair.
[linearly-scaled, growth of $1]
And the numbers…
The large vs small-cap seasonality map has been as successful as the long/short map (if not more), and the higher win % has resulted in a more consistent equity curve. There isn’t a whole lot I think I need to fix here.
In short…
I’m pleased (not elated, but pleased) with the performance of the Seasonality Map so far. Weak predictions on the long/short Map clearly need to be reviewed.
Beyond that, I think the real-time results above demonstrate that, though seasonality doesn’t by itself justify a trade, it is consistent (and significant) enough to justify being a tool in the trader’s toolbox.
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: Random Stuff | 3 Comments
Hi Michael,
since “strong” prediction days have done well, have You tried to be short/long only on days >= something (say 50%) and in cash on the others. What are the results in that case ??
Thank You
Marco
Hello Marco – results would have been very strong. Trading only days at least +/-50%:
54%, 1.4x, 15.6% (same 3 statistics as the post: win %, W/L, vol-adj return)
At least +/- 75%:
55%, 1.6x, 24.2%
At least +/- 100%:
63%, 1.4x, 32.1%
Just those pesky +/- 25% predictions dragging the numbers down.
michael