Do widely-used CMAs predict future performance?

Capital market assumptions — expected returns across equities, bonds, credit, and alternatives — are a core input into asset allocation decisions. Large investment houses and consultancies publish their estimates annually, and in many cases make them freely available to investors. For investment committees with limited time and resources, they are a natural starting point: they come from big names in the investment world, appear to be carefully constructed, and they cost nothing. But there is a question worth asking: how well do they actually work?

In this piece, we evaluate the track record of freely available CMAs against subsequently realised returns, explain what drives the performance gap we find, and draw out what this implies for how institutional investors should approach capital market assumptions going forward.

CMAs need to signal future returns

Even the most carefully constructed expected return estimates will not match subsequently realised returns precisely. Markets are subject to shocks – events change cash flows and prices away from what was previously expected - and CMAs are not meant to fully anticipate these in advance.

The more tractable question is whether a set of expected returns tells us about the ordering of future returns across assets. An asset class or market segment with a higher expected return than alternatives should, on average, deliver higher realised returns, even if the exact magnitude differs from the forecast. Getting the ordering right is integral to any portfolio construction: any optimisation procedure will favour assets with higher expected returns relative to alternatives, adjusted for risk. Strategic asset allocation is sensitive to the relative ordering of expected returns, not just to the level of individual estimates.

A powerful approach for evaluating CMAs is therefore to compare realised and expected returns across assets -- for example a scatter plot with expected returns on the x-axis and subsequently realised returns on the y-axis. A positive correlation indicates that higher expected returns were associated with higher realised returns — i.e. the ranking was broadly correct. A negative correlation means the opposite: investors would have been systematically better off doing the reverse of what the CMAs suggested. Recent research has documented that CMAs are indeed relied upon for asset allocation decisions, shaping the composition of huge amounts of capital across institutional portfolios worldwide.

How freely available CMAs have performed

We collected all available historical data from six major providers of freely available CMAs — drawn primarily from large investment houses and spanning the period from 2013 to 2026. We focus on equity markets, where forecasting returns is challenging and where getting the ordering wrong in terms of asset allocations has the largest portfolio consequences. We use five- to ten-year non-overlapping windows of returns to avoid the statistical pitfalls of overlapping observations.

The scatter plot below shows the results across all six providers. Expected returns on the x-axis, subsequently realised returns on the y-axis. The negative correlation is striking: providers that assigned higher expected returns to a market segment subsequently saw lower realised returns from that segment. Investors following these CMAs would have systematically underweighted the markets that outperformed.

Expected returns from six CMA providers, vintages 2013–2020, evaluated at each provider's native horizon (5Y, 7Y, or 10Y) where complete through 2025. Each point is one provider–segment–vintage observation.

An obvious contributor to this pattern is the persistent outperformance of US equities over the past decade and more. Most freely available CMAs assigned US equities relatively low expected returns over this period — constrained by elevated starting valuations and mean-reversion assumptions — while the market continued to deliver well above what valuations implied. When we remove US equities from the sample, however, the pattern does not reverse. The correlation between expected and realised returns sits near to zero: outside of the US, the freely available CMAs are essentially uninformative about the relative attractiveness of equity segments.

The fact that all six providers made similar errors is telling. It points toward a common methodological approach — reliance on historical return averages and valuation-ratio mean-reversion — rather than independent failures. When providers use slightly different versions of the same model, they share the same blind spots.

For comparison, we show the scatterplot with our estimates below. Unsurprisingly, the predictions are not perfectly aligned with the realisations but they get the ordering approximately right.

Expected returns at 5-year horizon, realised returns measured over subsequent 5-year window. Data are non-overlapping. US/Europe Equities covers markets, segments, and sectors within the respective regions.

Why are we able to deliver better CMAs?

The large investment houses do not lack analytical resources when building CMAs. The issue is that standard CMA methodologies embed an assumption that valuations revert to historical averages. That has not held in practice, and historical return averages are not the best guide for future returns in markets that undergo structural change.

The most consequential miss over our evaluation period was the relative return on US equities versus the rest of the world. The consensus view, implicit in most freely available CMAs, was that elevated US equity valuations signalled low future returns. What this framing missed was that higher valuations can be justified by a better long-term growth outlook — and that elevated valuation ratios can co-exist with sizeable expected returns for extended periods if the underlying fundamentals support them.

The case for higher long-run earnings growth in US equities relative to other developed markets has been reinforced over time by a better growth outlook that could have been captured in real time. Freely available CMAs, anchored to valuation ratios and historical averages, were not built to capture this.

Our own approach does not rely on mean-reversion in valuation ratios. Instead, we use a forward-looking present-value framework that extracts market-implied expectations for cash flow growth, inflation, real rates, and risk premiums at each point in time, and uses these to construct expected returns that are consistent with current market pricing. The chart below compares our estimates of five-year expected returns (nominal, local currency) for US (green) and euro area (blue) equities, illustrating how our approach tracked the widening differential between the two markets.

We started producing real-time CMAs in 2024, and our estimates could have had the benefit of hindsight. We address this issue in a few ways. Our estimates use forward-looking data that is available in real time, and we do not estimate any in-sample parameters to guide mean-reversion in returns. In addition, we have shown out-of-sample evidence that our short-term expected returns predict future returns in the period since we began producing estimates.

The current environment asks more of capital market assumptions than freely available estimates are built to provide. Besides getting the ordering wrong, the existing CMAs are too aggregated, too infrequent, and too confined to long horizons. Investors need expected returns and risk estimates that are granular, timely, and integrated with scenario frameworks to navigate the current environment. Freely available CMAs fall short, and the institutions that produce them have other priorities. Our CMA offering is designed to fill this gap.