Taking investors’ views into account when building portfolios is difficult but essential, says Roderick Molenaar.
A key use of our Expected Returns publication is to build portfolios that are aligned with our views in order to achieve the best investment results. As investors often use input from different sources to optimize their portfolios, translating their views on all the respective assets is no mean feat.
In particular, one needs to define point estimates for each asset class, as well as have a similar confidence in these estimates. For example, do investors have the same level of confidence in their view on interest rates as they do in their view on stocks? What if they do not have a view on all the assets in the investment universe? And what, for example, would be the best course of action if one asset class were expected to outperform another by a certain degree?
These questions are not new, and unsurprisingly, a great deal of work and research has gone into developing toolkits to help answer them. A major breakthrough came in the early 1990s, when Fischer Black and Robert Litterman wrote two papers introducing a methodology for portfolio allocation that explicitly takes into account investors’ confidence or uncertainty with respect to their own views of markets.
The approach is based on the recognition that, unlike in physics, there is no certainty when it comes to the financial markets. For example, to estimate the excess return of equities over the risk-free rate, we can use a historical average. Using realized returns for the MSCI World Index, starting in 1969, the average annual excess return equals 5.2%. Given the length of the data set, this seems like a good neutral approximation of what to expect in terms of excess return.
However, there is always a chance that the average has been calculated using a sample that does not fully represent the true characteristics. The standard error of the mean is 2.5%. Therefore, we can say that our expected average annual excess return for equities will likely lie between 2.7% and 7.7%.
However, in practice the 5.2% point estimate is often used without taking the accompanying uncertainty into account. The Black-Litterman model addresses these flaws and tries to improve on them using a three-step approach.
Step 1 – derive the implied views: The logical starting point for any portfolio construction approach is the benchmark. A benchmark should fit an investor’s objectives, and often represents the optimal long-term asset allocation. The first step in the Black-Litterman model is therefore to derive the ‘implied’ views from the benchmark.
These ‘implied’ returns justify the benchmark weighting of each class. If the returns do not match the investor’s long-term expectations, this should have consequences for the benchmark. This part of the model is in itself quite useful for most investors.
Step 2 – input the actual views: In the second step, investors give their views on (a selection of) the assets. Besides the views, the level of confidence in the views must be input for the model. If confidence in a certain view is low, the specific view will have a low impact on asset allocation in the portfolio; if it is high, it may have a significant impact.
For portfolio construction, this approach is very appealing as it eliminates the need to input views in a certain format; views can be provided on not only the absolute but also the relative performance of assets. Besides it is sufficient to provide views only on a subset of the assets.
Step 3 – combine the views: Having calculated the views implied by the benchmark and our actual views, we combine them. For this, we use what is known as a ‘Bayesian framework’. This is a statistical toolkit that enables us to determine the return views based on implied views and actual quantitative views. These combined views will subsequently be the input for the optimization approach.
The Black-Litterman model requires us to input point estimates of our expected returns and the level of confidence assigned to those views. In practice, these requirements can reduce the effectiveness of the approach, especially when it is hard to summarize the information into the point estimates.
Therefore, we have enhanced the model to make it more accessible. Often strategists or investment committees first rank assets either in relative terms, e.g. ‘asset A will outperform asset B’, or in absolute terms, e.g. ‘rates will increase more than forwards’. Transforming these qualitative views into more quantitative ones, e.g. ‘asset A will outperform asset B by 2.5%’, and also quantifying the confidence investors have in the views can be challenging.
An approach that eliminates the need to translate qualitative views into quantitative ones is therefore desirable. We believe that this additional step enables investors to better express their views in their asset allocation.