In delegated portfolio management clients trust the economic expectations and investment acumen of their asset managers. The asset manager’s task is to maximize the clients’ portfolio return while at the same time fulfilling the details and requirements contractually agreed upon.
Clients want to understand the reasoning why their assets are invested in a specific allocation. Such transparency is equally important when machines make autonomous decisions (e.g. robo-advisors). However, the much-touted transparency that comes along with digital asset management is still rather vague when it comes to explanations of underlying capital market views.
How an asset allocation is formed is the proprietary knowledge and skill of investment managers. However, all resulting asset allocations have in common that – given numerous specific restrictions and constraints – the allocation is believed to maximize achievable returns. In other words, each asset allocation clearly expresses the market expectations and beliefs of its creators. For a comparable level of risk, different asset managers allocate widely varying assets and proportions.
When it comes to robo-advisors, things get even more difficult. Providers of passive digital asset management services rely on static (or strategic) asset allocations. Client portfolios then are periodically rebalanced towards this strategic asset allocation. Providers of active autonomous asset management models typically claim that their asset allocation decisions are purely based on statistical processes, highlighting that this eliminates the subjectivity of human decisions.
A weakness of both static and statistical approaches is the impossibility to explain investment decisions, particularly in the context of temporary economic effects. E.g., clients may read about worries over increasing trade barriers affecting global economic confidence, but there is no logical connection to the allocation in their portfolio. For statistical allocations there may be a statistical explanation, depending on a lot of parameters. However, this will most likely not create any explanatory value for the customer.
When it comes to robo-advisors, many claim that they hold no market views. This is false. We outline the two most common approaches versions and our own approach.
Common approach for passive robo-advisors: static portfolio weights with rebalancing
The simplest approach taken by most available robo-advisors is to construct a number of sample allocations with different risk levels and apply these static weights to all customers choosing a particular risk strategy. To prevent too much divergence from this allocation, portfolios are rebalanced regularly, e.g. quarterly, by selling some of the better performing assets and investing the proceeds into the underperformers.
This implies that if a particular allocation is advised by an asset manager at all times, the market view regarding the relative potential of asset classes and the diversification of the assets technically always stays the same. Moreover, while effective risk of such an allocation may diverge substantially from one month to the other, the asset manager still advises to change nothing. Consequently, clients who buy passive robo advice services basically receive a sleek (often one-off) online way to invest money cheaply without much economic market view and research in the background.
The robo advice comparisons using real money conducted by brokervergleich.de record for portfolios of ca. 50% equity a two-month divergence of +1.6% (Solidvest) to -0.2% (easyfolio). For portfolios with ca. 60% equity a divergence of +2.1% (Ginmon) to -2.8% (Cominvest) between May and June of 2018 is reported.
Different market views must have shaped the portfolios. The way they are formed is largely opaque but the emphasis on rebalancing means that the views are cemented rather than adjusted to a differently viewed world.
The economic argument for this relies on mean reversion. Recent underperformers are now “cheap” and may therefore rise again soon, outperformers are now “expensive” and further growth is relatively more difficult, so gains are realized and transferred to the “losers” of the past. This means that, in any market condition, the biggest loser is bought. However, fundamental explanation or economic backup for decisions are virtually impossible.
Common approach for active Robo-Advisors: statistical asset allocation
Statistical models provide another way of generating return expectations for automated asset management. Companies using this method usually claim that their choice of a suitable model reflects their deep understanding of financial markets, but the chosen parameterization massively influences the outcome. Transferred to classical advice, such an asset manager advises on a particular allocation given the contractual agreement, thus she must believe that the recommended allocation reflects her views best – or totally forfeit economic competence.
Statistical models thus do represent views but must rely only on historic data and a model, such as drift/ momentum, long term averages or linear regressions. The choice of data and parameterization is essential for the outcome.
Imagine this straightforward example: the mean return of a set of assets calibrated over a period of 20 years is almost always significantly positive due to the survivor bias. Over a ten-year period, the same mean return is still almost always positive but sometimes less than the 20 years mean return since one prolonged crisis may have a large impact. Thus, the model only deals with positive return expectations for all assets. Over a 5, 3, or 1-year time frame it may fluctuate widely from +30% to -20%, indicating that some assets have negative return expectations and should under no circumstances be in the portfolio.
Simply applying the client’s planned investment horizon to the data thus would produce wildly diverging portfolios with strongly contradictory implied market views to explain to clients. To make things worse, return expectations derived from historical data tends to be pro-cyclic rather than anti-cyclic.
More complex statistical models suffer from the same truth; the choice of the model produces the return expectation. A 12% allocation to US small caps is not adequately explained by the inclusion of a size factor in a regression model, it actually means “we as company X believe that in the next year they are going to appreciate”
Portfolio construction using any statistical model (with the notable exception of pure risk contribution models such as Risk Parity) tries to achieve the best return for a given risk by evaluating combinations of assets. When it is done in automated fashion by a robot rather than a human, it relies on pre-set models that produce non-interpretable and highly fluctuating numeric return expectations.
A better approach: Strategic expert allocation with statistical models
Human experts use economic reasons for decision making and can explain using logic and causality, even if occasionally they turn out to be misguided. However, we are convinced, that well-informed humans still have a very important role to play. Understanding broader trends and especially political events is where the complex world is better interpreted with expert insight. Whether a political referendum has lasting impact on the state of a union or how a previously unimaginable economic policy such as negative interest rates may impact markets cannot be handled adequately by historic data extrapolation or regression fitting alone.
Therefore, we apply a proprietary approach to transforming human insight into machine-readable return expectations. In a regular asset liability committee allocation committee (ALCO) comprised of capital markets experts, for which we conduct research and present analyses, we decide on a strategic asset allocation for a wide range of asset classes. The strategic allocation is guided by a risk corridor determined by an objective benchmark.
Importantly, the economic projections and market views of the ALCO change over time and are forward-looking. The macroeconomic outlook of various regions, geopolitical situation and stock market valuations are assessed with a perspective of six months to one year. Expectations regarding foreign exchange rates also factor into the allocation with hedged and unhedged options. New or niche investment opportunities (such as ESG stocks, Private Equity or Water) may be included if the ALCO has a positive view on these assets.
The allocation agreed upon at times can be more bullish or more bearish than its defined benchmark. The ratio of the risks of ALCO and benchmark allocations indicate the current risk appetite of the committee and represents a key parameter for determining the absolute return expectations for each of the considered asset classes.
These return expectations are calculated on a daily basis, using a proprietary algorithm to derive the relative return expectations implied by the reference allocation given the risk and diversification structure of the asset classes. This also allows to calculate implied values for assets which are not currently allocated; e.g. when US stocks with currency risk are allocated, US stocks with hedged FX-risk may after all still have a positive return expectation but not the most profitable trade-off at this point.
These (relative) return expectations are then translated into absolute ones and can be used to construct portfolios of any risk level. This ensures consistency of all portfolios but allows for sufficient flexibility to accommodate all types of individualization.
As outlined before, any asset allocation has return expectations that must hold for this allocation to be “optimal” by the considerations of the asset manager given the existing market dependency structure.
We call these return expectations “implied returns” since they are derived from our strategic asset allocation. Here is a fictious example showing the effects when the view on USD FX market changes. Imagine the investment committee decided to shift a 9% allocation from US Large Cap Equity to US Large Cap Equity EUR hedged. The US stock market may still be vibrant, but market dynamics and recent currency appreciation may have changed its consensus opinion on the exchange rate path.