How to avoid extrapolation bias
“The difficulty lies not so much in developing new ideas as in escaping from old ones.” ― John Maynard Keynes
Past performance may not be indicative of future results, say the regulatory disclaimers. Whichever way you approach asset allocation it relies to some extent on the past. And the past is a long time.
As economists have observed, one of our key human weaknesses is the tendency to overweight the recent past in any analysis of the future. In his book, Against the Gods, Peter Bernstein, explains what is commonly referred to as Extrapolation Bias.
“So we pour in data from the past to fuel the decision-making mechanisms created by our models. But therein lies the logician’s trap: past data from real life constitute a sequence of events rather than a set of independent observations, which is what the laws of probability demand…[…]. It is in those outliers and imperfections that the wildness lurks.”
In reality, all most of us have known is one, more or less, continuous trend, namely, falling bond yields and increasing bond prices. So, however one thinks about asset allocation, this fact and various associated cross-asset-class correlations are baked in.
The dangers of statistical extrapolations are highlighted by two famous stories: ‘When Elvis died in 1977, 170 people impersonated him professionally. In 2000 that figure had grown to 85,000. Extrapolating, experts would predict that a third of the world’s population will be Elvis impersonators by 2019’, according to an article on trends¹.
The second is the Wisdom of Horse Manure² and the Great Horse Manure Crisis of 1894. ‘By the late 1800s, over 50,000 horses were transporting people around London. On average 1.25 million pounds of horse manure was deposited on the streets each day. The Times newspaper predicted in 1894 that in 50 years’ time, every street in London will be buried under nine feet of manure.’
A neat demonstration of computer science terminology, ‘garbage in, garbage out’.
The smallest change in expected return inputs can lead to out-sized imbalances in asset allocation. In Finding the right asset allocation – Is optimisation the solution?, Towers Watson gives the example of a four asset class (fixed income & equity) portfolio, in which an imperceptible change in 10 year real return assumptions for Gilts (from -0.8% to -0.6%) resulted in the proposed Gilts weighting dropping from 16.6% to zero.
Written in March 2003 (years before the global financial crisis), in Are Portfolios Obsolete? (which is referenced in Investment Insights), Bernstein called into question the validity of policy portfolios, which are designed to be a strategic baseline mix of asset classes.
The key flaw Bernstein saw with policy portfolios was that they are based on long-term assumptions of the future. “In a volatile world, opportunities and risks will appear and disappear in short order. Flexibility is the watchword,” he said.
In a review of Edgar Lawrence Smith’s 1924 book, Common Stocks as Long Term Investments, John Maynard Keynes warned, “it is dangerous to apply to the future inductive arguments based on past experience, unless one can distinguish the broad reasons why past experience was what it was”.
So, again quoting Keynes, how do you avoid falling into the trap of expecting the results in the future that will materialise only if conditions are exactly the same as they were in the past? Perhaps a better way to think about asset allocation is to target asset and investment class risk rather than asset class returns.
By this we mean turning the conventional notion of a portfolio allocated to 60% equities 40% bonds (and later an allocation to alternatives to diversify), on its head. In what can be defined as the third generation of asset allocation, portfolios are now constructed around ‘drivers of returns’.
Lionel Martinelli, director of the EDHEC-Risk Institute, agrees. He believes that the investment management industry has entered a new paradigm where standard strategic asset allocation portfolios are dead, liability driven investing is in vogue, and in parallel to this, there is the emergence of ‘factor investing’.
Factor investing, also known as risk premia investing, are terms used to describe the process that isolates these drivers of returns. It allows allocations to be made in terms of risk factors as opposed to investments in a number of different asset classes (avoiding the inherent risk of having to forecast the future).
This way managers can change the way they do business from a market-centric point of view focusing on investment products, to providing investment solutions from an investor-centric perspective.
The merits of this approach are reduced margin of error and the ability to pre-plan portfolio re-balancing in the event of a changed risk scenario, such as a crisis or market liquidity shock.
Some institutional investors have already replaced capital budgeting with risk budgeting, for some parts of their portfolio. For AP2, risk premia was the way to access hedge fund ‘returns’ without the fees, while Norges Bank Investment Management has been using it as an equity beta enhancement. Denmark’s PKA has adopted factor investing to get exposure to specific risk-premia and ‘market effects’ in equities.
If this does prove to be the next iteration of asset allocation, risk premia will become the way to allocate across the whole portfolio. As AP2’s chief investment strategist says³, “We want to get paid for taking systematic risk, not pay for it.”
Whatever the outcome for politics, for bonds and for inflation over the next 20 years, it would seem that the time for Bernstein’s call for flexibility is now. And risk premia looks like the way to do it.
¹ Richard Tomkins: A theory on trends (6.6.2005), Financial Times
² The wisdom of horse manure (2.9.2013), Financial Times
³ Investors learn to harvest hedge fund returns sources without high fees (22.4.2012), Financial Times
Photo: © Niki Natarajan 2017