Past performance is not indicative of future results: using ’forward-looking’ estimates of expected returns

We have previously highlighted issues that come up when using historical returns exclusively to define Capital Market Assumptions. In this article, we explain in more detail how we construct ‘forward-looking’ expected returns.

A key question for investors is how they should interpret and use historical return and valuation data for thinking about the future. A basic answer to this question – common in practice - is that an asset’s future returns can be gauged by taking an average of historical returns. A more nuanced answer – also commonly used across the industry - is that future returns can be gauged by taking averages of return components i.e. historical earnings growth and/or reversion of valuation ratios to historical averages.

Despite drawbacks, the use of historical returns and average valuations to estimate expected returns is widespread. This is understandable to an extent: taking historical averages has some statistical basis and allows investors to avoid modelling the underlying economic structure that generates returns. It is also easy to communicate estimates that are closely tied to historical performance.

While we agree that an expected return methodology should be easy to communicate and intuitive, this should not come at too high a cost. We think that a forward-looking approach to estimating expected returns can remain intuitive while making better use of historical returns and data.

What do we mean by forward-looking expected returns?

When estimating expected returns across asset classes and investment horizons, we use forward-looking inputs wherever possible. Our present-value approach gives clear guidelines on how market expectations of cash flows and risk premiums can be translated into expected returns. Forward-looking expected returns are constructed using forward-looking data: options, futures, swaps, and professional surveys, that capture key components of macro present-value relationships.

Our macro present-value framework ensures that expected return estimates align with market-implied discount rates, cash flow expectations, and market prices at each point in time. Expected return estimates that are not consistent with market pricing – such as those based on historical returns - are effectively off-consensus predictions rather than estimates of expected returns. Such estimates are implicitly taking active macro views relative to what is priced into markets.

For strategic asset allocation, it is more appropriate to think in terms of expected returns on passive asset class exposures. As a basic illustration of this idea, we can get a simple read on expected returns on government bonds by looking at their yields. This measures expected returns assuming that the bond cash flows come in as expected by the market. This is an entirely forward-looking assessment, and there is no reason to expect that expected returns based on historical data will deliver this. The same logic applies to other asset classes.

While we minimise reliance on historical returns and valuation ratios, this does not mean that we disregard historical data. We still need to extract insights from history that inform our modelling of expected returns. In particular, historical data are needed to understand the underlying economic structure that generates asset returns. This structure can then be used to model expected returns across a much wider set of assets when forward-looking derivatives or survey data are not available. We will outline this extended approach to expected returns in a future Insights piece.

Why a forward-looking approach matters

Timely and consistent estimates of expected returns

The benefits of forward-looking estimates are clear when it comes to evaluating expected returns in real time, especially in volatile periods when expected returns can change abruptly. Our use of options and futures in present value models means that expected returns adjust immediately to re-pricing of market expectations. The estimates are therefore unimpeded by historical averages or similar that, by definition, only update slowly.

The benefits of our approach are further illustrated using the ‘Value’ component of expected equity returns. This component is driven by the difference in estimates of short- and long-horizon expected returns, both of which are estimated daily using forward-looking data and consistent with current market pricing. Many CMA approaches estimate the Value component by assuming mean-reversion in valuation ratios to some historical average. These ratios are often calculated based on averages of earnings or similar over a long period of time, meaning that they are slow to update. In addition, variation in these ratios is not necessarily connected to transitory variation in expected returns, as they can be driven by other factors such as changes to the long-term outlook for earnings growth.

An natural test of expected return estimates, especially at shorter horizons, is to evaluate forecasting performance in a backtest. We plan to discuss the predictive power of our estimates of expected returns in more detail in a future Insights piece.

Estimating expected returns on government bonds

A key driver of the expected returns on government bonds is the term premium, especially at shorter horizons. The traditional way of estimating expected returns on government bonds is to either estimate the term premium using a model of historical yield curve dynamics or to use the yield curve slope as a noisy proxy for the term premium.

Both approaches implicitly assume that the future will look similar to the past, though in different ways. Using the distance of the current level of yield curve slope from its historical average implicitly assumes that there is a stable level to which the slope will always revert. We have discussed in the past the major issues with extrapolating from past bond performance when the yield curve embeds persistent macro trends. In addition, it is very difficult to obtain accurate estimates of the term premium without accounting for macro trends in yields.

Accounting for macro trends is best done in a forward-looking way, closely approximating the information set investors have in real time. To implement this, we estimate the long-term inflation expectations and r-star using forward-looking data – yields, inflation derivatives, professional surveys and Fed dot-plots. The term premium is estimated in a term structure model jointly with the macro trends. As a result, our estimates of the term premium and expected returns on government bonds are always aligned with the current market pricing and more accurate than the alternative approaches.

The engine that implements forward-looking estimates of expected returns is a core part of our ASAMM framework. Estimates of expected returns and risk for a broad range of asset classes are available via our Capital Market Assumptions solution.