Drilling Down on the TAA Model’s Real-time Performance

01May12

My personal trading is focused in two places right now: (a) trading volatility ETPs, and (b) tactical asset allocation (TAA).

My vol. trading is a big return generator today, but I recognize that I’m exploiting market imbalances that may not exist long-term.

On the other hand, TAA isn’t going to light the world on fire, but I can rest well knowing that we’re chasing opportunities that have existed for a century and more.

The TAA model has underperformed its benchmark in the 19 mos since inception.

As I’ve shown before, the model would historically have gone through long periods of underperforming, especially when the benchmark has been strong. This is a “generational” model, and I’m much more concerned with returns over the next decades than any one month or year.

But I think it’s still a worthwhile exercise to drill down on where the model has done well and not so well, and where it could be better.

Things the Model has Done Well

I’ve held U.S. Treasuries (IEF) and Gold (GLD) for most of the last 19 months. Those trades have done well – so well that ALL of the model’s returns since inception have come from just those two asset classes.

Things the Model has NOT Done Well

The model has done a poor job at both selecting specific equity asset classes to invest in AND timing the % of the portfolio to invest in equities each month.

Note that when I say equity asset classes I include real estate (VNQ) and commodities (GSG), because in today’s market, all are highly correlated.

To illustrate, in the graph below I’ve shown two portfolios. The first (grey) are the model’s actual “equity-driven” returns. The second (blue) are the results of simply investing in the volatility-adjusted equivalent of the S&P 500 (SPY) each month.

What I’m testing is the model’s success in picking specific equity asset classes (grey), versus just dumping each month’s equity exposure in to the S&P 500 (blue).

[growth of $1, linearly-scaled]

The fact that the grey line consistently trails the blue means that the model has done a bad job selecting equity asset classes.

A bit of that is due to bad luck (ex. taking a big position in EWJ right before the Tohoku earthquake), but most of it is bad timing.

But the suck doesn’t stop there.

Recall the blue line in the graph above, which is a different % of the portfolio invested in SPY each month. What if I just took the average exposure of all months, and invested it on day one. Here I’m testing the model’s success timing the % of the portfolio to invest in equities each month.

[growth of $1, linearly-scaled]

The fact that the blue line (slightly) trails the red means that the model has also done a bad job timing the % of the portfolio to invest in equities.

Next Steps

The first of these shortcomings (selecting the specific asset class) is an issue I need to continue analyzing. The model would have been better off just plowing equity exposure into the S&P 500 most months so far and that’s unacceptable.

The second shortcoming (timing the equity exposure) I chalk up to “stuff happens”. Like Mebane Faber (the inspiration for my own model) I’m using long-term moving averages to guide entry/exits in to equities, and that’s a concept that’s served investors well for a century plus (given enough time to let the averages play out).

I’m at my self-imposed word limit. More perhaps in a future post.

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   |  7 Comments