Future Now
The IFTF Blog
John Kay on financial models
John Kay's recent piece in the Financial Times is a useful warning to people doing financial—and other kinds of—forecasting: understanding the historical limitations of your model, and how badly things can go when the assumptions built into your model no longer hold:
Quantitative portfolio management relies on measures of correlations between asset classes. These historical correlations are not universal constants but the products of particular economic conditions. Unless you understand the behaviour that produced them, you cannot assess their durability. In 2007-08, assets that had been uncorrelated were strongly correlated and many portfolio managers were surprised when the diversification they sought proved illusory.
Underlying causal relations had changed, as they frequently do in business. In the new economy bubble of the 1990s, equities roared ahead while property languished. But during 2003-2008, the availability of underpriced credit, followed by its abrupt withdrawal, affected property and shares in similar ways. Anyone in the financial world knew these things: but computers, churning through reams of data, did not.
Asset classifications change their meaning. The alternative asset classes that yielded strong returns in the 1990s for Harvard and Yale were hedge funds and private equity. But the increase in the number of hedge funds and the volume of their assets meant that an investment in the sector—once a bet on an individual’s idiosyncratic skills—became more similar to a general investment fund. Hedge fund returns were therefore increasingly correlated with those of other investments.
Private equity was once a punt on small entrepreneurs. A manager with good judgment could make money from a few hits in a diversified portfolio. But by 2006 the sheer size of private-equity funds led them to focus on well established businesses. Such investments were a geared play on the stock market. They no longer spread risk: they concentrated it. Worse, many uncorrelated assets appeared uncorrelated in the past only because they were thinly traded and infrequently valued. Pressures to “mark to market” revealed the underlying correlations.