4

Shorter-Term Macro Signals and Something About Driving a Truck

Finance has borrowed a lot from macroeconomics.

—JPP

WE’VE SEEN THAT ALMOST ALL VALUATION SIGNALS WORK reasonably well across markets, perhaps more than most investors realize. Correlations between equity valuation signals and forward returns almost always have the expected sign. In fixed income markets, yield ratios seem to be remarkably predictive.

I wouldn’t make the statement that investors can make a lot of money if they mechanically follow simple valuation-based strategies, but as we’ll see in Chapter 12, systematic strategies that combine valuation with momentum signals have worked well in backtests and in practice. As for nonsystematic processes, when I described our own approach to TAA, I explained that we look at a variety of factors, not just relative valuations. Again, it helps to think in terms of building blocks. Relative valuations are key building blocks, which we can add to other factors.

For example, there is a wide body of academic literature that suggests macro factors can be significant drivers of asset returns. And among practitioners, statements such as “Stocks make money in expansions and tend to lose money in recessions” are often held as self-evident. My view is that short-term return forecasts and tactical asset allocation decisions must account for macro factors. However, there is little published on how to use these factors to inform investment decisions. In a 2017 paper titled “Macroeconomic Dashboards for Tactical Asset Allocation,” my colleagues David Clewell, Chris Faulkner-MacDonagh, David Giroux, Charles Shriver, and I take the practitioner’s perspective. We show how to build dashboards to integrate macro factors into a broader, discretionary TAA process. Our goal is not to design stand-alone systematic trading strategies based on macro factors. Rather, we show how investors can build macro factor dashboards to introduce discipline into their asset allocation process (in combination with other inputs, such as relative valuations).

In Chapter 3, I showed that valuation signals don’t always have very high correlation with forward returns. One of the reasons could be that valuation- based investment strategies tend to be more effective when valuations are at extreme levels. Importantly, strategies that focus solely on relative valuations can lead to disappointing outcomes when important macroeconomic shifts take place. There’s ample evidence that macro factors also matter.

Academic Research on Macro Factors

Most of the academic literature focuses on whether macro factors get priced into markets. Chen, Roll, and Ross (1986) show that the sensitivities (“macro betas”) of size-sorted stock portfolios to rates, industrial production, inflation, credit spreads, and consumption explain a significant portion of their relative performance over time. Fama and French (1989) use a different methodology that focuses on the broad stock and bond markets. They show that business conditions, as approximated by dividend yields, rates, and credit spreads, forecast broad market returns. Several other studies have confirmed the importance of macro factors in explaining a wide range of asset class and style premium returns. Factors covered in the literature include consumption, unemployment, inflation, GDP growth, and oil prices. I highlight some examples in Table 4.1.

TABLE 4.1 Prior Studies on the Influence of Macroeconomic Factors on Asset Returns

Challenges with Practical Applications

While these studies provide credible evidence of the importance of macro factors, many practitioners still struggle to use these factors for tactical investment decisions at the 6- to 18-month horizons. Economists and investment teams often operate independently, and the question of what macro expectations are priced in markets is often left unanswered. Moreover, the sheer amount of macro data makes it difficult to separate noise from signal and anticipate which variables will drive returns.

Another challenge with prior studies in practice is that macro factors may influence asset class returns differently based on initial conditions. Boyd, Hu, and Jagannathan (2005), for example, show that a rise in unemployment has a different effect on stock returns if it occurs during an expansion or a recession. Similarly, we can expect that a decline in industrial production may have a different effect if starting business conditions are good or bad. In fact, in our paper we suggested that the same could be said of any macro factor—depending on the prevailing regime, the impact on asset returns will differ. Yet except for Boyd, Hu, and Jagannathan’s study, previous research does not account for the relationship between current conditions and the subsequent impact of macro factors on asset returns.

To map macro factors to expected asset returns, we proposed using dashboards. We now use these dashboards in our Asset Allocation Committee. The approach is different from the econometric methods used in academic studies—it is meant to be simpler and more intuitive. In fact, one of my colleagues referred to these dashboards as “paintings by numbers.” There’s more that can be done to extract signals from macro data, and our research team continues to innovate in that area. But as I’ve mentioned a few times in this book, simple is good. Simple works.

Unlike historical regression analyses based on static data samples, our dashboards are meant to be dynamically updated such that investors can rely on them as a research tool or as a means to inform investment decisions on an ongoing basis. We focus on the relative returns between pairs of asset classes. We highlight which factors may have a significant impact on which pair trades, under various scenarios. Importantly, we incorporate current conditions, as reflected in the macro factors’ levels. Our dataset covers a broad list of key macro factors: industrial production, inflation, oil prices, spreads, rates, gold prices, unemployment, etc. These factors are mapped to the pair trades we’ve discussed in this book so far (stocks versus bonds, value versus growth stocks, say small cap versus large cap, high yield versus investment-grade bonds, etc.).

For each pair trade, we partition historical asset returns to match a given scenario and current conditions. In total, we map 18 pair trades to 10 key macro factors, for a total of 180 “cells,” produced for each of 4 scenarios (full sample, as well as stable, rising, and declining macro factors)—a grand total of 720 cells. We use color coding to make it easier to scan the data and mine for trade ideas.

Our entire framework is out-of-sample. Starting from each macro factor’s current level, our dashboards answer the following question: If an investor has a one-year view on the direction of the macro factor, what is the corresponding forward one-year return?¹ For example, recently inflation has been low. Should we expect that value stocks outperform growth stocks if inflation rises from its current level? To answer these types of questions, we report historical hit rates, average returns, and confidence intervals across the 720 “cells.” It’s a lot of data, but we can easily spot some patterns.

In our paper, for example, we studied the link between US large cap versus small cap stock returns and the USD:

As of April 10, 2017, the U.S. dollar index stands at 100.6, which is in the top quartile of its history since January 1990. Suppose a tactical asset allocator expects the U.S. dollar to rise further. Starting from a top quartile level, when the U.S. dollar subsequently rose by 5% (or more) over the next year, U.S. small caps outperformed U.S. large caps 88% of the time, by an average of 8.2%, with the 10th to 90th percentile range between –2.3% and 15.9%. The outperformance of U.S. small caps in periods of rising USD may be attributed to their lower reliance on exports, compared to U.S. large caps.

This example seems significant, but we also explained that in general, statistical significance between macro variables and asset returns is low, especially if we don’t expect a significant move in the macro factors. When I first presented this framework to our Asset Allocation Committee, I wanted to be clear that, despite all the academic literature in support of the significance of macro factors, most of these numbers should be ignored. You could have driven a truck through these confidence intervals with room on either side.

We should use the dashboards in context, not in isolation. Investors should think about what’s already priced in, look for extremes—cases where we see real historical significance—and put the analysis in the context of current conditions (valuations and other nonmacro factors).

The bottom line is that these dashboards help identify relevant investment themes. Here are additional examples from our paper:

Emerging markets currencies may be an important factor to watch. Stable or rising emerging markets currencies are supportive of emerging markets equities, real asset equities, and emerging markets bonds. Further, emerging markets currencies have depreciated significantly—the index currently sits at the bottom 5% of its historical range. If they move significantly up or down from their currently low level, they could correlate with meaningful directional volatility across assets. And the price of oil, if it remains stable or appreciates from its currently medium level (63rd percentile), could be a significant positive driver of emerging markets stocks, real asset equities, and emerging markets bonds. . . .

Regarding style rotation, the direction of interest rates should matter in the current environment. Growth stocks have longer duration than value stocks. Therefore, even though dividend yield on value stocks is higher than on growth stocks, when rates decline, growth outperforms; and when rates rise, value outperforms. This effect occurs both in the United States and in EAFE markets. The large weight of negative- duration Financials in the value index partly explains this effect.

Overall, the results reported in our dashboards are in line with economic intuition, as well as with prior findings published in the literature. The contribution lies not in the dashboards’ academic merit, but rather in their value to practitioners. The confidence intervals and hit rates help filter the continuous flood of macro data. Importantly, while the relationships among the macro factors and asset classes are reasonably persistent, the dashboards should be updated frequently, because as initial conditions change, some of the investment conclusions may change as well.

Caveats

With our macro dashboards, we don’t claim to identify causation, which is almost always impossible to do given the complex and dynamic nature of how factors drive asset returns. Rather, we merely identify correlations that appear meaningful and leave it to the investor to assess causation. Investors shouldn’t build systematic tactical asset allocation strategies based solely on these macro data.

Instead, macro factors are often used to confirm relative valuation signals. For example, if non-US equities are cheap relative to US equities based on valuation metrics (P/E and other such metrics) and if macro factors indicate non-US equities should outperform (weakening USD, non-US central bank stimulus, earlier stage in the business cycle versus that in the United States, potential GDP growth surprise, and so forth), then a tactical asset allocator may take a larger position in ACWI ex-US equities than if valuation and macro data don’t agree.

Another caveat is that we don’t model expectations directly. In theory, we should run our scenarios against the expectations that are priced into the market. The problem is that expectations are often difficult and, in many cases, impossible to measure. Survey data may be useful, but they rarely reveal what markets are truly pricing in, nor are survey results available in a timely—or broad enough—fashion going far enough back in time.

Regarding market-implied views, forward curves incorporate a risk premium, which makes it hard to disentangle an expectations component. Ultimately, Chen, Roll, and Ross (1986) conclude that spreads and interest rates series are noisy enough to be treated as unanticipated. The authors also find that econometric methods to extract the unanticipated component of industrial production do not offer any advantage over the unadjusted series.²

Last, we’ve selected investable asset class pairs. This list represents asset classes that asset allocators commonly use in practice. But in theory, it would be more elegant to isolate market factors and scale positions based on volatility. For example, we could hedge the equity risk factor common on both sides of the small versus large caps pair, or at least make sure the trade is equity-beta neutral. While statistical significance would likely increase (see Naik et al., 2016), the trades would not be easily implementable. Ultimately, our goal is to add discipline to the analysis of macro factors, and our framework is one piece of the puzzle, focused on idea generation. Portfolio managers can then combine macro with other factors, adjust broad market factor exposures, and risk-scale positions between the long and the short leg and across trades.

Notes

1.   Here by “forward return,” I mean the expected return if the macro factor scenario is realized over the next year.

2.   However, they lead industrial production by one year. For tactical asset allocation, this obviously would be like cheating because it would assume perfect foresight.