The Decision Moose Asset Allocation Framework

A reader suggested a review of the Decision Moose asset allocation framework of William Dirlam. “Decision Moose is an automated framework for making intermediate-term investment decisions.” Decision Moose focuses on asset class momentum, as augmented by monetary policy, exchange rate and interest rate indicators. Its signals tell followers when to switch from one index fund to another among nine encompassing a broad range of asset classes, including equity indexes for several regions of the globe. The trading system is a long-only approach that allocates 100% of funds to the index “having the highest probability of price appreciation.” The site includes a history of switch recommendations since the end of August 1996, with gross performance. To evaluate Decision Moose, we assume that the 69 switches and associated trading returns are as described (out of sample, not backtested) and compare the returns to those for the dividend-adjusted S&P 500 Depository Receipts (SPY) over the same intervals. Using data for the 72 trades spanning 8/30/96 through 9/21/12 (16 years), we find that:

In calculating SPY total returns by Decision Moose trading interval, we assume trades occur at the close on Decision Moose signal dates, or at the close on the next trading day if signal dates are not trading days.

The following table summarizes gross Decision Moose trading results over the entire sample period (72 trades) and over the five years ending 9/21/12 (28 trades). Over the entire sample period (last five years), Decision Moose:

• Signals an average of about 4.5 (5.6) trading decisions per year.
• Generates a gross profit for 81% (68%) of trades.
• Outperforms buying and holding SPY by an average 2.7% (0.6%) return per trade.
• Outperforms buying and holding SPY during 54% (46%) of trading intervals.

Including trading frictions would shave the outperformance of Decision Moose by a small percentage depending on specific broker fees, fund bid-ask spread and trader account size. For small accounts, this friction may have been important.

Is the gross outperformance of Decision Moose relative to SPY persistent over time?

The following chart summarizes the differences between Decision Moose (Moose) gross returns and SPY returns by trade over the entire sample period, with a dashed best-fit linear trend line. The trend line indicates that Decision Moose outperformance may have dissipated over the sample period.

The sample size of 72 trades is not large. Might the Trade 18 outlier (gold during 11/24/01-6/1/02) be decisive in determining the trend?

The following chart summarizes the differences between Decision Moose gross returns and SPY returns by trade over the entire sample period, excluding the Trade 18 outlier. The best-fit linear trend line still indicates dissipation of Decision Moose outperformance over time, though not as severely as above.

The trend in outperformance by trade could be misleading because recent Decision Moose trading frequency is higher for the recent subperiod than for the overall sample period. What happens to the trend if we normalize trading results based on outperformance per calendar day?

The following chart summarizes the differences between Decision Moose gross returns and SPY returns per calendar day by trade over the entire sample period. After this normalization, Trade 18 is no longer an obvious outlier. In fact, Trade 51 (avoiding much of the Fall 2008 crash) is the best normalized interval of outperformance.

The best-fit linear trend line indicates dissipation of Decision Moose normalized outperformance over time.

A plausible interpretation of these dissipation tests is that financial markets are adapting to increasing use of momentum-based tactical asset class allocation strategies. However, sample size is not large, and differences in outperformance over subperiods could be random.

What does a regression show about the relationship between Decision Moose gross returns and SPY returns?

The following scatter plot presents a regression-based perspective on normalized gross Decision Moose performance over the entire sample period. Results indicate that:

• Decision Moose generates a gross daily alpha of about 0.02% relative to buying and holding SPY. (However, over the past five years, this alpha is about -0.01%.)
• Decision Moose has a beta of about 0.31 relative to buying and holding SPY.
• SPY returns explain about 12% of Decision Moose returns (the R-squared statistic is about 0.12).

The latter two points indicate that Decision Moose returns are only loosely linked to U.S. stock market performance.

In his FAQs, William Dirlam suggests that Decision Moose trading is best suited to reasonably large tax-deferred accounts to minimize the impacts of trading friction and taxes. He also offers guidance on the type of investor for whom Decision Moose is suitable.

In summary, the Decision Moose asset allocation framework may offer investors a way to beat buying and holding the broad U.S. stock market by occasionally trading to the “hottest hand” (in economic context) from a set of nine asset class proxies, but its outperformance may be dissipating.

Cautions regarding findings include:

• As noted, the sample of trades is not large for trend analysis.
• As noted, Decision Moose may not work well with small accounts for which taxes are not deferred.
• SPY may not be the most appropriate benchmark for Decision Moose, which employs nine distinct asset classes.