In other words, our model relies on back tests, but these back tests are run over extremely long periods of time. Many of our competitors use 3 or 5 years of historical data to back-test models, while we use 3 or 5 decades. That’s the only way we can be sure that we’ve met both bull and bear markets, both strong and weak economic growth periods, volatile and stable environments, a few market crashes, and a few bubbles.

Running back tests over such a long period of time can be a challenge because for the 1970s or the 1960s, when nothing was computerized, it is more difficult to find reliable data. So we spend a lot of time rebuilding proprietary databases containing macroeconomic, company and market information, to make sure that they are clean, perfectly accurate, and reliable.

The research cycle itself starts with back testing of the way each individual indicator is working. If we want to know if the strategy based on finding low P/E shares is valid in the US, we would check how the indicator performed in the very long term. If it performed well, we would divide the period into sub-periods to check how this indicator works under all types of environments. We would measure its effectiveness, compare it, and would get into more and more detail about its reliability and consistency.

Gradually, we eliminate indicators that do not seem satisfactory. Then we work on different ways of combining the good ones properly. Once we’ve found the right combination, we work on how it can be used to actually build a portfolio; from which level the stock should be considered a sell if it goes down in the ranking, from which level it can be considered a buy if it goes up in the ranking. We also determine how many stocks we should have in the portfolio, how we should weight them, how often we should rebalance the portfolio, etc.

Q:  What are the general rules that you follow in terms of portfolio construction? What’s the turnover of the fund?

A: The portfolio construction rules are simple. We hold 50 stocks that are equally weighted. They have a target weight of 2% and we let them deviate by 0.35% before we reweight them. Of course, we have tested much more sophisticated solutions but, with the indicators we had selected, none of them appeared to work better than equal-weighting.

As for turnover, it has varied over time depending on market environment but the number of stocks replaced in any given year has been between 26% and 58%. It is relatively low but I should point our long-term focus. We still have a few stocks I never bought myself, although I’ve been managing this strategy for more than 10 years. Sometimes we buy a stock and sell it a month later if something has changed, but the typical time of a stock in our portfolio is in excess of 2 years.

Q:  How do you monitor the risk and what risk control measures do you have?

A: This is probably one of the most differentiating points in our approach.

Risk management is undertaken primarily by built-in control mechanisms introduced as early as the development stage of the quantitative model. The backtests that are carried out to build a well-performing, stable strategy involve an indepth evaluation of the model’s structural risk characteristics under all kinds of environments.

Although our model does not attempt to quantify each portfolio's ex-ante tracking- error explicitly, it has been tuned to generate a relatively stable risk over time, thanks to the complementary contributions of its stock selection principles. The most commonly used systems estimate and manage risk on the basis of an assumption that each asset's individual risk profile can be assessed through an analysis of its past evolutions. We tend to doubt the reliability of ex-ante tracking error calculations produced by such systems because, at each company's level, too many exceptional events can occur that would make this assumption completely wrong. In the approach we have adopted, the only hypothesis regarding the future is that the behavioral patterns of market participants, which we identify through backtests, will not suddenly undergo significant changes. This assumption, in our view, is much more acceptable, because any structural, durable change in the behavior of the millions of investors who form the market would only take place slowly, and thus be detected by our continuous research effort.

Other more explicit factors, such as the maximum sector deviations from the index and the number and weight of portfolio holdings are only marginal additions to the risk control process. The judgmental overlay also plays a role in limiting risk, as its detailed assessment of the company's outlook primarily aims to identify any bias or weakness that may impede a proper evaluation of all the potential downside by the model.

Moreover, the fund management team uses several other tools to monitor the various aspects of the risk involved in their portfolio. In particular, an internal """"""""dashboard"""""""" providing numerous comparisons to evaluate the differences between the portfolio’s profile and that of the index is produced every month. Finally, the team runs the Barra Aegis and UBS PAS tools, to help analyze risk from a different standpoint.