When I wrote the Devil’s Derivatives I looked at the risk modelling behind the ratings of CDOs, bank trading and credit portfolios. Through a combination of bad incentives and leverage, these models performed very badly and contributed to the financial meltdown of 2007-8. Lurking in the background of my analysis was traditional portfolio modelling as used by investment managers, which was plucked off the shelf and then misused by investment bankers and credit raters.

I sometimes wondered how that kind of modelling was used in a context without leverage or flawed incentives such as ratings-based capital requirements. Over the past couple of years I got some insight into this in a job where I met colleagues and clients who were either building risk models for investing purposes or using them in their capacity as heads of risk at buyside asset management companies. It struck me that a quiet revolution was in progress.

These models take a universe of stocks and bonds and break down their historical performance into specific factors based on characteristics such as style, geography or industry. Having done that, you can take a particular fund or portfolio and the model tells you how much of its performance and risk comes from specific factors, the underlying index or idiosyncratic stock-picking not captured by the model.

As these models become more commoditised and widely available, the community of people using them has grown. Even if a traditionalist active fund manager sticks with a qualitative, narrative-based approach to investing, someone out there is using a model to check on their performance. That person might be an in-house head of risk and compliance, or a pension fund or insurance company that allocated money to the manager and wants to check that they are staying within their mandate and not overcharging for the performance they deliver.

With these models increasingly commoditised and widely used, fund managers are starting to use them proactively. Rather than use the factors to break down historical performance of existing funds, managers might consider a particular factor and try to design a new fund or exchange-traded product that captures its performance, using optimisation and back-testing software to refine things as they go along.

To see how this thinking works, consider the following chart of weekly style factor returns produced by the Bloomberg global equity portfolio model.  Going back to 2005, this chart shows the striking outperformance of value as an investment style with an annualised Sharpe ratio of over six, in contrast to the weak performance of growth. While it would be quite hard to capture that risk-adjusted return in a long-only fund (because of inevitable exposure to the volatile index), I suspect that some long-short equity hedge funds must have come close. I would love to know who they were.

After I generated this chart a couple of months ago I sent it to a contact, Michael Kollo at Renaissance Asset Management. He immediately fired back with some charts of his own, generated with the help of Bloomberg’s arch-rival in the modelling space, MSCI Barra.

What MSCI Barra did was upgrade its model, breaking down the single value factor into two components – a price-to-book factor and an earnings yield factor. It turns out that since 2007, the superior returns came from the second of these, and not the first. Presumably, armed with this information, MSCI Barra clients might go out and launch new hedge funds or ETFs based on that insight.

Of course, fund managers will tell you some caveats apply. While an optimiser will tell you how to capture the return and risk profile you want, in practice it might be impractical to deliver because of transaction costs and short-selling constraints (particularly in emerging markets).

Other fund managers will warn you not to get too transfixed by the output of a model lest you lose sight of the world around you. For example, Barra’s model shows that the residual volatility factor (tracking stocks that were particularly volatile irrespective of the underlying index) would have lost you 28 percent since mid-2007. However, that could simply be a side effect of post-crisis deleveraging driving investors into low-volatility stocks.

What does seem undeniable to me is that the growth of factor modelling has made fund management much more quant-based than it used to be. That explains the growth of ‘smart beta’ where factors or weighting schemes are embedded into new low-cost passive products. Even active managers are presenting themselves in a more quantitative way.

For example, Kollo recently launched a new fund focused on high-dividend emerging market stocks (hint: buy Saudi Arabian utilities). With its focus on factor returns and back-testing, the prospectus resembles a smart beta product even though the fund is actively managed. While in banking, risk models are seen as part of the problem of complexity and too-big-to-fail, in fund management, they’re increasingly part of the woodwork.