In our 2015 paper, ‘Hypercube in the Kitchen: Reading a Menu of Active Investment Strategies’1 by Gnedenko and Yelnik, we showed how asset managers may be classified based on their skills. Our classification could be viewed as a five-dimensional hypercube. In this discussion we focus on one of its edges: discretionary versus systematic.
The systematic style represents a combination of a systematic strategy (a.k.a. systematic model)’s decisions, and those of the portfolio manager. Whilst the model’s trades are implemented by changing positions in financial instruments, the systematic manager’s trades are changes made to the systematic strategy itself.
Evaluation through statistical methods of investment managers, be they discretionary or systematic, is only possible within certain limits.
Similarities between discretionary and systematic investment management firms, in terms of their exposure to the key person risk, are stronger than is commonly believed to be the case.
Trade and its limits to statistical evaluation of managers
Understanding the differences between discretionary and systematic managers begins with an understanding of trade: for a discretionary manager, a trade is a one-time activity in buying or selling financial instruments. For a systematic manager, buying or selling financial instruments is the model’s trade, not his or her own. The systematic manager’s equivalent is making a change to the model.
When a discretionary manager makes his or her trade, the decision is based on conviction (which is, in turn, based on analysis, experience, mood, etc.). Expected return of any one trade of a well performing discretionary manager is only marginally positive. It is impossible to statistically reject the hypothesis that such return is zero, since every trade is unique and not repeatable.
How do investors identify a good manager in practice, leaving qualitative assessments aside? One of the primary measures is the manager’s Sharpe ratio. T-value of returns may be estimated as the Sharpe ratio multiplied by the square root of time, assuming the manager’s returns are independent and identically distributed (i.i.d.), which rarely happens in real life. With a Sharpe ratio of 1, a four-year track would give us a t-value of 2, a value that allows us to say the manager’s returns were statistically different from zero (with a confidence of 95 per cent). With a very respectable Sharpe ratio of 0.7, one would need to wait more than eight years to achieve this level. Nobody does this in practice, not least because there is no reason to believe that a discretionary manager’s returns will remain static over such a long period of time. As a result, when selecting discretionary managers, investors largely ignore statistical performance indicators. Their call is predominantly based on the qualitative skill of identifying good discretionary managers (ironically, this skill is hard to quantify, for the same reason).
A discretionary manager relies on the ‘black box’ of a human brain. A systematic strategy (though not a systematic manager) appears remarkably similar to that of a discretionary manager from the outside. The model’s trades are no different from a discretionary manager’s individual trades. Trying to estimate the statistical significance of either is futile. This degree of similarity is high enough to suggest that a good systematic strategy, having been put on an autopilot for a reasonable time, will deliver results on par with those of a good discretionary manager.
However, inside a systematic strategy, returns may decompose into individual constituents. These constituents are not trades in financial instruments. They may be called ‘trading rules’: factors, indicators, risk management techniques, weighting schemes, etc. Trading rules are what drive trades in financial instruments. Since this happens regularly, we can often – but by no means always – collect sufficient statistical data to reach conclusions about trading rules’ statistical significance.
Looking at back-tests helps systematic managers better understand ideas. A trade made by the systematic manager is a change in constituents of the systematic model. When the systematic manager selects a trade (i.e., introduces changes to trading rules), they typically prefer it to be statistically significant, but must that be the case? We would argue that the answer is ‘not necessarily’. To start with, a systematic manager’s trades are conceptually similar to the trades of discretionary manager or a systematic model, but are made in a different trading universe, i.e., in constituents of the systematic model. On top of that, even if a constituent showed a statistically significant performance in the past (or, especially, in a back-test) it would require a leap of faith to assume that the constituent could be at least as successful in the future. In other words, the systematic manager’s trades – much like the discretionary manager’s trades – may be based on conviction.
How often does it make sense for a systematic manager to ‘trade’? Intuitively, a systematic manager’s time in their trade should be greater than their model’s time in its trade (or, at least operate at the same order of magnitude; an exception may be a stop-loss trade in which it quickly becomes obvious that a new constituent is not fit for purpose).
Typical investors do not distinguish between the systematic manager’s trades and the model’s trades, as the realised track record encompasses both. In the absence of a sufficiently long record, though, investors often assume that the model’s trades represent both well enough. In so doing, they ignore any shifts introduced by the systematic manager’s trades into the stream of the model’s returns.
An obvious question arises about separability of the systematic manager’s returns and the model’s returns; however we are not aware of any studies on this subject.
With regard to statistical analysis of returns, investors typically assume that systematic managers look after their models constantly, have the chance to change them every day, and therefore have as many data points as there are days or months in their sample. In practice, systematic managers typically introduce only infrequent and marginal changes to their models. This causes the effective number of trades to be small, which makes statistical studies of systematic managers’ skills less trustworthy. However, investors prefer healthy conservatism to frequent model changes – or, in other words, prefer the absence of style drift to the ability to statistically verify the systematic manager’s skill (which would be difficult regardless, due to scarcity and opacity of the systematic manager’s trades).
Key person risk and institutionalisation
Intuitively, people dislike black boxes. A computer program is traditionally perceived to be more opaque than the human mind. Systematic managers are pushed to institutionalise their process because, unlike in discretionary management, the process can be said to be less dependent on any one individual. In other words, the managers promote it as a trade-off, whereby a reduction of key person risk comes at the expense of increasing perceived complexity and, by implication, opacity of strategy.
In considering related trade-offs, it may also be noted that:
Institutionalisation reduces the transition period in both discretionary and systematic management companies2, and means more people are involved in building the strategy and therefore have a good knowledge of it. In a discretionary management firm, this implies a move towards a process-driven approach and independence from the intuition of one key portfolio manager. What is the trade-off? Involving more people invariably leads to greater consensus thinking. Investors often deem this too high a price to pay when selecting a discretionary manager, as they tend to prefer to buy into an individual. However, in selecting their systematic manager, investors prefer to buy into a process, and are therefore prepared to accept the price of institutionalisation. In response to investors’ demands, systematic management companies take on their share of the price increase by hiring more research staff. Contribution to the systematic manager’s trades (i.e., the content of the systematic model) becomes increasingly marginal as the number of bright, well educated and well paid people increases. These effects of institutionalisation of the systematic management companies are very similar to those observed among their discretionary peers.
Both systematic and discretionary managers seek to capture risk premiums and inefficiencies. Analysis of these managers invariably considers their classification along the discretionary versus systematic edge of the hypercube of skills. Whilst the two styles have significant differences, they are similar in some respects. These similarities are especially striking as they relate to statistical evaluation of a manager’s skill, and the nature of the key person risk.
1. Available at http://papers.ssrn.com/abstract_id=2536408.
2. Discretionary companies also institutionalise their investment processes. Many managers have formalised processes in place even though final trading decisions are still discretionary.