Many pension plans use risk systems to test how different scenarios would affect their portfolios.
Traditionally, risk models use thousands of scenarios. The likelihood that these scenarios will occur is weighted equally and generally reflects short-term asset characteristics. On the other hand, pension plans also use asset-liability models for long-term portfolio construction. This could result in a mismatch between the time horizons used for risk management testing and those used for asset allocation decisions.
But there may be a better way for long-term institutional investors to conduct risk analysis — by assigning different weightings to scenarios based on their likelihood of occurring and the longer-term timeframe, according to a new working paper by Jacky Lee, vice-president of investment analytics at the Healthcare of Ontario Pension Plan, Redouane Elkamhi an academic at the University of Toronto and Sheikh Sadik a student at U of T.
“The aim of this article is to propose a simple methodology on how a risk system, while calibrated with a shorter modelling period, can be leveraged appropriately to be more [aligned] with long-term strategic asset allocation,” said the paper.
It proposes a methodology that re-weights various scenarios based on their likelihood, which is inspired by change of measure and importance sampling techniques. “The main point of this paper is that we show it is possible to impose different asset characteristics by simply re-weighting the scenarios,” says Lee.
An example is a scenario in a risk system where the S&P 500 index is down 20 per cent and bonds are up five per cent. This would be a reasonable scenario when looking over the past 10 years, he says. “But if you actually look at more long-term history — so let’s say back to [the] 1970s — stock bond correlation has actually been around zero. If you have this scenario where the S&P is down 20 per cent and [bonds are] up five per cent, we can say that this scenario is more likely to come from a short-term history than a long-term history.”
So if a risk system was calibrated using the short-term history, but a pension plan wanted to reflect the long-term history, it could underweight the influence of the scenario that’s more likely to come from the short term, says Lee.
Most asset managers have some kind of risk system or asset liability modelling system that can leverage this methodology, notes Lee. “I think . . . one of the key insights of our methodology is, once you use the system to generate a lot of scenarios, most asset managers will use the simulation to calculate value at risk metrics, expected tail loss [and] standard deviation, but they are equally weighting scenarios because the model assumed certain characteristics and wanted to use them.
“What we are contributing to the literature is that, by reweighting those scenarios, you can actually alter their characteristics. For example, change the correlation between stocks and bonds, and what have you, so that you can calculate risk metrics based on another set of assumptions by simply reweighting the scenario of your existing system without actually running the system again or changing the settings or getting a new system.”
As technology becomes more advanced and risk systems mature, vendors will be looking to gain an advantage by providing more insight with their tools, says Lee. “I feel like, if they implement some of the capabilities outlined in this paper, then perhaps the asset management industry might adopt this approach as an additional tool in their tool set to manage risk and to provide the capability to align the assumptions of the tool and the assumptions they want to make.”