Tactical Asset Allocation For The Real World
Managing risk via tactical asset allocation (TAA) offers a number of encouraging paths for limiting the hefty drawdowns that take a toll on buy-and-hold strategies. But what looks good on paper can get ugly in the real world. There’s a relatively easy fix, of course: consider the total number of trades associated with a strategy as another dimension of risk. The dirty little secret is that many TAA backtests don’t survive the smell test after considering the impact of trading frictions – particularly for taxable accounts. Deciding where to draw the line for separating the practical from the ridiculous varies, based on the usual lineup of factors – an investor’s risk tolerance, time horizon, tax bracket, etc. But there’s an obvious place to start the analysis. Let’s kick the tires for some perspective using some toy examples. An obvious way to begin is by using the widely cited TAA model outlined by Meb Faber in what’s become a staple in the literature for this corner of finance – “A Quantitative Approach to Tactical Asset Allocation.” The original 2007 paper studied the results of applying a simple system of moving averages across asset classes. The impressive results are generated by a model that compares the current end-of-month price to a 10-month average. If the end-of-month price is above the 10-month average, buy or continue to hold the asset. Otherwise, sell or hold cash for the asset’s share of the portfolio. The result? A remarkably strong return for the Faber TAA model over decades, in both absolute and risk-adjusted terms, vs. buying and holding the same mix of assets. The question is whether running the Faber model as presented would be practical after deducting trading costs and any taxable consequences? Let’s ask the same question for two other simple strategies: Percentile strategy: apply the rules in Faber but limit the buy/hold signal so that it only applies when the asset price is above the 70th percentile for the ratio of the price above the trailing 10-month average. The same logic applies in reverse for the sell signal: the asset price is below the 30th percentile for the ratio of price below the 10-month moving average. For signals between that 30th-70th percentile range, the previous signal remains in force. Relative-strength strategy: apply the Faber rules but limit the buys to assets in the top half of the performance results for the target securities, based on the trailing 10-month results. The same rule applies in reverse for triggering a sell signal. In other words, sell only assets in the bottom half of the performance results via the trailing 10-month period if a sell signal applies . Note that for all strategies, the signals are lagged by one month to avoid look-ahead bias. To test the strategies, we’ll use the following portfolio (see table below), which consists of 11 funds representing a global mix of assets, spanning US and foreign stocks, bonds, REITs and commodities. In essence, this is a global twist on the standard 60%/40% US stock/bond mix. The initial investment date is the close of 2004 with results running through this month as of Feb. 26. All the models start with the same allocation. The chart below compares the results for the three strategies and a buy-and-hold portfolio. The Faber model delivers the best results. A $1 investment in the strategy at 2004’s close was worth roughly $1.48 as of last Friday. The Relative Strength model was in second place at $1.39, followed by the Percentile Strategy ($1.34) and Buy and Hold ($1.20). Raw performance data tells us that the Faber model is the winner. Note, too, that all three TAA models deliver superior results in risk-adjusted terms. For instance, historical drawdown for the three strategies is relatively light compared with the Buy and Hold model. In particular, the Buy and Hold portfolio suffers a hefty drawdown in excess of 40% in 2008-2009 whereas the three TAA models never venture below a roughly 10% drawdown. Given what we know so far, it appears that the Faber model is the superior strategy via a mix of strong performance and limited drawdown risk. But the results look quite a bit different once we add in the dimension of total trades associated with each strategy. Buy and Hold, of course, excels on this front. But the lack of trades (or trading costs) is more than offset by the steep drawdown for Buy and Hold. The question, then, is what is the superior TAA model if we consider real-world costs? The numbers provide the answer via a summary of total trades for each strategy, as shown in the table below. The Percentile model’s trades number just 93 for the 2004-2016 test period – less than half the trades for other two strategies. The Percentile model’s total return trails the Faber results, but only modestly so. In short, the Percentile model generates 90% of the Faber model’s returns, with a comparable level of superior drawdown risk compared with Buy and Hold. Add in the Percentile’s substantially lower turnover clinches the deal, or so one could argue. If this was an actual consulting project, we would run additional tests before making a final decision. For instance, we might consider other models and look at longer historical periods, perhaps using daily prices and compare results with a variety of risk metrics. Running Monte Carlo simulations to effectively test the models thousands of times would be useful too. Looking at the results in terms of the number of trades associated with each strategy is no less valuable. This subtle but crucial aspect of backtesting tends to be ignored. But if you’re comparing TAA models for use in the real world, it’s essential to adjust for real-world trading frictions. In some cases, adding this extra layer of analysis may end up as a determining factor for separating failure from success.