Our Developed Markets ex-US suite of models (AXDMxUS4) was designed to better address the risk management challenges of model users with benchmarks, such as the STOXX® Global 1800 ex-US. By excluding the dominant influence of the US in risk and factor return calculations, we believe managers can at times obtain more accurate risk forecasts, especially in active space, compared with using the Worldwide (AXWW4) model.
Focus on Factor Returns
Over the long run, the factor returns from the two models are closely related, as most developed markets tend to behave similarly. Rolling one-year correlations of five-day factor returns are generally high and relatively stable (Exchange Rate Sensitivity is the obvious exception, Figure 1). Over shorter periods, however, returns can be quite different and sometimes significant.
Figure 1. Rolling One-Year Correlations of Five-Day Returns, Worldwide vs. Developed ex-US Models
Take, for example, 2020 (please). For some factors, the elimination of the US in the investment universe meant that returns were very different between AXWW4 and AXDMxUS4 (Figure 2). Most notable was the big difference in the above-noted Exchange Rate Sensitivity factor, where the AXDMxUS4 return was very positive, but the AXWW4 return was equally negative. Although this factor measures a stock’s sensitivity to movements in its home currency (and the raw value is therefore the same for both models), the high degree of flip-flopping values in the US as the dollar strengthened sharply and then reverted – and the corresponding hugely negative return to the factor in the US – meant the factor fared far better when the US was excluded from the universe.
Conversely, Market Sensitivity had a much higher return in AXWW4 than it did in AXDMxUS4. Although returns were similar in both models when the market dropped, investors’ risk-seeking behavior was stronger in the US as the market began to recover, especially over the summer. Note, too, that the correlation of five-day (and daily) returns remained high this year. The return difference was driven by a much higher magnitude of returns in the US, but they did move in the same direction.
A few other factors also saw large gaps in returns of the same factor across the two models, and in most cases the differences persisted throughout the year. Dividend Yield, Leverage, Liquidity, Value and Volatility all saw higher returns in AXWW4. No factors aside from Exchange Rate Sensitivity showed consistently higher returns in AXDMxUS4.
Why is this important?
Both attribution and portfolio construction are impacted. If you have an exposure to one of these factors in your developed markets ex-US portfolio, having such a different return may very well distort the results when AXWW4 is used for attribution. Assume a relatively small positive Market Sensitivity exposure of 0.1. Attribution from AXDMxUS4 would show the factor contributed only 7 basis points to your return, whereas AXWW4 would show a contribution of more than 50 basis points. Since you could not invest in the US, you could not possibly achieve the higher return from AXWW4. That difference would be likely to go into the “stock specific” category when it is not stock specific at all, rather it is just a vestige of the risk model used, and certainly not a measure of your stock-picking ability.
The difference in returns also factors into the future volatility forecast for the factor. AXWW4 may “think” Market Sensitivity has higher volatility, since it experienced higher daily returns in that universe. If you are using AXWW4 for your risk-managed developed markets ex-US portfolio construction, you may use more of your risk budget on your exposure to that factor, which may prevent you from using it on another factor, potentially hurting returns from other sources.
In many periods, portfolio managers can create perfectly good developed markets ex-US portfolios using a worldwide model that includes the US, because a model with US stocks does not always have much of an impact on risk. However, in certain periods, such as 2020, a manager would probably have been better off using the model more specific to the mandate, both for more accurate risk management and better explanatory power of performance attribution.
Figure 2. 2020 Cumulative Factor Returns and Return Differences, Worldwide vs. Developed ex-US Models
 As with all our other models, we offer two horizons, short- and medium- and two factor structures, fundamental and statistical. This analysis is based on the medium-horizon fundamental model.
 With apologies to Henny Youngman.
 The AXDMxUS4 model also eliminates emerging markets, of course, but the US typically has more influence on factor returns.
 We introduced the concept of the impact of constraints, such as investment universe, no-shorting and reduced trading frequency in our paper “What, Exactly is a Factor”. This is another example of the importance of aligning – or at least understanding the impact of – your attribution model with your investment process.