Recently on a trip to Cape Town, South Africa to speak at the IMN African Cup of Investment Management conference, I had a long discussion with one of FactSet’s client relationship managers, Andre Benjamin, and a client, about what steps to take in choosing a risk model? Considering we have hundreds of risk models on FactSet from five major risk vendors, it’s become increasingly important to help our clients choose a risk model.
First, the five vendors available on FactSet are R-Squared, Northfield, SUNGARD-APT, Axioma, and MSCI-Barra. These vendors offer a variety of models with local or regional specifications, varying update frequencies and choices of either statistical or parametric (pre-specified multi-factor) models. Additionally, multi-asset class (MAC) risk estimating capabilities are also available with R-Squared/FactSet, Barra, APT and Northfield.
If you’re a long-only UCITS large-cap equity manager residing in Paris, and, due to losses suffered during the Credit Crisis of 2008, you want to update a risk management process while implementing an active risk management strategy, which model is best for you? Or if you’re a long/short global equity manager in Hong Kong who occasionally invests in corporates, which product should you choose to reveal where likely losses may occur from under stress testing scenarios?
These questions and more have prompted me to create a list of questions whose answers will serve to make the risk model selection process easier.
Outline the investment process. You want to identify those considerations most important to you before you choose a solution. In this way, you are properly defining the problem you’re attempting to solve and not getting ahead of yourself.
Reacquaint yourself with your investment policy statement or guidelines. The major investment risks are written there in black and white for you and will lead you toward ruling out risk models that are not capable of shedding much light on those types of risks. This is not entirely essential for the model selection process and may not offer any clues, but it’s a good place to start. At a minimum, it will reacquaint you with what the stated objectives your investment process is trying to accomplish, because it’s not uncommon to forget why they were written and what’s in them.
Investigate prior risk management strategies you or your firm have tried. It’s important to understand what has worked in the past and what hasn’t. There’s no sense in choosing a model which mimics what you have not had success with formerly.
Discern whether global or regional models are best, whether statistical or factor models are most appropriate, and whether daily or monthly updating frequency models are most suited for your investment process. Understand that there is no “right” model, there’s only the model which you’ll feel most comfortable with because the model’s recipe and/or factors are more closely aligned with your priorities and risk objectives.
The more you trust a model, the more you’ll use it. The more you use it, the greater the identification of mechanisms for losses, the lower the volatility, and the lower risk your portfolio will become over time. It’s better to use a less optimal risk model more often than a “best fit” risk model less often, and being comfortable with a risk model is a priority for its usefulness.
With these questions researched and documented ,it’s time to ask yourself more direct questions. Before doing so, identify who will be using or running the risk model and risk analysis and include them in the conversation.
Direct questions, not in any particular order involve:
What type of assets do you invest in? What are your comparative indexes?
What key factors are involved in your security selection process for investing?
Regional, country or global investing?
Are you interested in explanatory variables only so that reporting risk is most important to you or will you use the risk reports for help in asset allocation? Asked another way, will the model become an integrated part of the investment process?
Will you perform exposure analysis or do you prefer just attribution by active weighting?
How often will you be running risk analysis?
What is your risk horizon?
What’s the annual portfolio turnover in a year?
Will you be including stress-testing?
Are you going to run portfolio optimization with the risk model?
Are you interested in hedging? If so what instruments or assets will you hedge with? What risk measures will you keying in on?
Are you primarily interested in explanatory models and reporting or forecasting and assisting portfolio construction and asset allocation processes?
Given this list you should be able to examine the factors in your narrowed selection of risk models to explore which of them have the factors you feel most closely match your investment process, assuming you’re not using a statistical model. You should optimally select a model whose estimation universe comes from the country or region you invest in.
The updating frequency of your mainstay risk model should be determined from your annual turnover in a year combined with how often you rebalance. Likewise the risk horizon can be chosen from this information. Other things to consider are the simultaneous use of a long-term and short-term horizon model. If the longer horizon model is your mainstay, by also monitoring your shorter term risks (specifically common risk versus specific risks as percentage of total), you’ll know whether your next rebalancing with the long horizon model should be to tighter or looser tracking errors, for example.
At this point, determine which portfolio(s) you’ll use for testing, preferably with at least one year’s history. Run a risk report displaying ex-ante risk forecasts and ex-post risk measures for these risk models of similar type over say, the last 12 months for instance. Along with this, an exposure analysis should be examined minimally at the industry level and comparisons between what the portfolio manager estimated his or her exposures to be versus what they really are should be performed. A useful activity is comparing active weighting with active exposure to observe the mismatch.
Additionally, consider using one model in portfolio optimization or with aiding asset allocation and another in which to view risk exposures for sanity checks. This practice can be quite illustrative. For example, a statistical model could be used in portfolio optimization and a parametric risk model for exposure analysis after the optimization is run.
Lastly, remember risk models demonstrate the trending of risks quite well, more so than the levels. Likewise, drilling down to more granular groupings to examine risk exposures and volatilities, eventually reaching the security level, disaggregates the averaging that reduced estimation error at the portfolio level, so be careful in their interpretation.