Capital market assumptions

By incorporating uncertainty we recognise that central expected returns for assets are estimated with error rather than assuming they are known, as is the case with mean variance techniques. A key benefit is that we can allow for different conviction levels in return expectations. We consider the distribution around the mean, effectively reducing the weight placed on our central estimate. Distinguishing between uncertainty and risk is important. We define uncertainty as the range of outcomes for the mean and risk as the range of outcomes around the mean. The amount of uncertainty we take into account for each asset classes depends on a number of criteria. They include the back-tested predictive power of our asset class return models, the historic volatility of assets and the desire for diverse portfolios when optimising.

Uncertainty in central returns feeds in to our stochastic simulations, that give a range of potential return pathways from five years out to the long term. When constructing portfolios, these simulated pathways and our central return uncertainty enable us to use robust optimisation techniques that generally lead to less concentrated portfolios compared with those portfolios resulting from mean variance optimisation. It also gives flexibility to focus on certain upside or downside scenarios when constructing portfolios to fit client needs. Read more.

We use Monte Carlo simulation to create random distributions informed by historical return distributions and centred on our expected returns. The engine simulates thousands of return pathways for each asset, representing the range of possible outcomes over a five-to 30-year time horizon. We leverage BlackRock’s risk models to ensure we respect co-dependencies between asset returns. The range of scenarios incorporate our work on incorporating uncertainty in return expectations. The Black-Littermanmodel (1990) –a well-known model for portfolio allocation -combines long-and medium-term views in a single-period setting. Our model uses a Kalman filter –an algorithm that extracts insights about potential future paths by bringing together a number of uncertain inputs -to extend this approach into a multi-period setting. This allows us to capture the variation of expected returns over time under various scenarios —from economy-related to market sentiment driven. A large part of these variations is not predictable. Constructing portfolios that are robust to, or can exploit, these variations is a major challenge for investors. The ability to calibrate the engine with asset class views with uncertainty at arbitrary time horizons, and to evolve this uncertainty stochastically, drives the dispersion of return outcomes. Highlighting the uncertainty that investors face when building portfolios helps ensure ostensibly precise return expectations do not lead investors to concentrated portfolios.

Simulated return paths support a broader range of applications, such as asset-liability modelling. Stochastically generated return scenarios enable investors to move with ease beyond mean-variance and optimise portfolios against their individual needs. Investors can place more emphasis on the tails of the distribution or focus on the path of returns rather than just the total return. They can incorporate flows in or out of the portfolio over the course of the investor’s time horizon or place more emphasis on scenarios that are challenging for the investor’s business beyond their portfolio. Investors with complex asset-liability matching requirements, such as insurers, typically rely on stochastic simulations of returns to assess and construct portfolios.

We put estimates of the equity risk premium (ERP) at the heart of our approach to setting return expectations. We calculate the equity risk premium using an implied cost of capital approach (Li et al, 2013). We use a discounted cashflow model and take today’s market price and expectations of future dividends and growth and interest rates to arrive at an implied equity risk premium.  Changes in equity valuations are driven by both expected cash flows – earnings and dividends – and the ERP. Forming expected returns by looking solely at valuations – typically the price-to-earnings ratio – can miss the full picture, in our view. Our work finds that linking expectations for future interest rates and the ERP can be more telling for expected returns rather than attempting to find a fair value for the price-to-earnings ratio alone. This allows us to incorporate our views of the structural drivers of interest rates into expected equity returns –as well as other asset class returns. We also use bottom-up analyst earnings forecasts and the relationship between margins and the economic cycle to formulate our earnings expectations (using an augmented discounted cash flow model). We find corporate profit margins not only converge to long-term averages but do so at a faster pace when an economy reaches full capacity. We assume in future the ERP will mean-revert to levels observed in the post-1995 period. Our projections for risk-free rate, or “long-run short rate”, are described in the Interest rates methodology section.

We derive our expected returns for government bonds by mapping out the yield curve at multiple time horizons in the future. This is based on estimating (1) the short rate, and (2) model implied term premia. Estimates of short rates are based on market data in the near-term and on macroeconomic informed data in the long-term. More specifically, in the long-term, we assume investor views about long-run inflation and real growth, coupled with changing preferences as to savings and risk aversion, will determine expectations for short rates (the “long run short rate”). Model implied term premia are computed from a model based in the affine term structure class of models (Adrian, Crump and Moench, 2013) describing the yield curve using the first five principal components of yield. The model implied term premia from the affine term structure model are further calibrated to market implied term premia, with the relative weights dependent on the relevant time horizon.

Our model for credit asset (excess) returns is anchored on two key elements: 1) our estimate of credit spreads at a given horizon and 2) our estimates of loss due to defaults and downgrades over the horizon. The first component is projected in a consistent manner with our view of real GDP growth, as implied by BlackRock’s factor-augmented vector autoregressive macroeconomic model (Bernanke, Boivin and Eliasz, 2004) and the link between credit spreads and equity volatility. Our approach attempts to avoid overfitting, yet retains the ability to explain a high proportion of the variance in credit spreads and passing cross-validation tests against more complex approaches. The second component is estimated based on our outlook for spreads, the duration of the asset and an assumed credit rating transition matrix which captures rating migrations and defaults across multiple credit cycles. We currently base our transition matrix on Moody’s long-run transition data. We aim to further develop our model by directly modelling transitions based on macroeconomic conditions in order to better capture cycle dynamics and the respective variation in losses due to credit events. In addition to making our estimates of credit spreads consistent with our macroeconomic views, the our new credit (excess) return model allows the flexibility of calibrating our expected returns to various credit rating compositions which may prevail over the entire time horizon.

The private market return models can be grouped into two categories – equity and debt. The equity models – relevant for core real estate and private equity buyouts – are based on an accounting statement framework. We estimate earnings growth and future valuations, which are used in conjunction with observable private and public market data (current valuations, financing cost, leverage, etc.) to model the evolution of the capital structure over time and infer equity returns. Estimated earnings growth and future valuations are linked to components of our public market return expectations for equity, rates, and credit spreads. Crucially, they also consider the unique dynamics of each asset class, such as the changing occupancy rates for real estate. Returns for private market debt – infrastructure debt and direct lending – are estimated using a ‘build up’ approach. The total return is a build-up of underlying public market factors (interest rates) and private-market specific return drivers such as credit spreads, losses due to default and downgrades, leverage and borrowing costs. Unlike most public debt markets, direct lending is modelled as a ‘buy and hold’ investment, in line with how investors access the asset classes. The published returns are gross of fees and we net out representative client fees, accounting for management fees, carried interest, and hurdle rates, when creating and optimizing portfolios.