In my previous blog, I outlined when it would be worth using a long-term risk model, and identified three fundamental reasons why you might need one. Taking a step back, how this applies in any situation will vary depending on what you’re trying to do. To frame the discussion, I shall use the example of a DB pension scheme.
For a DB pension scheme, the risk is being unable to pay the pensioners, and the point of a model is to make better investment decisions. There is a simple framework for doing this: using a deterministic model, solve for the returns required to reach investment targets (accounting for contributions), and then compare the risks of different portfolios with sufficient expected returns using a simple 1 year risk model (and other metrics – such as stress tests).
To compare this approach with the use of a long-term risk model, it is instructive to think about when the approaches would give fundamentally different answers. After all, the risk is not so much that the model is wrong, but that recommendations based on it are poor1. There are only two ways the results could differ- a long-term model might attribute a portfolio with higher or lower risk than the 1 year model.
If the longer-term model attributes more risk to a portfolio, it would imply that, in the model, the portfolio was too conservative to meet its objectives; so, while losses in any given year would be small, the portfolio would not pack enough punch.
However, to make this claim you would have to be quite confident in your return distributions, which means you should be at least as confident in your mean and/or median returns. In this case, this risk can be captured much more simply with the deterministic expected vs. required return framework. Since you’d only consider portfolios designed to meet the targets, this need not be a problem for the simple 1 year model.
Alternatively, the long-term model might attribute less risk to a portfolio than the short-term model. For this to be true, the long-term model would ultimately have to be saying that you could lose big in the short-term, but you’d recover it afterwards2.
The question then becomes, are you confident enough in your model to believe it when it says you can ride out large downsides? Do you know the assumptions well enough, and do you understand their interactions well enough, to trust that you can ride out heavy losses? Are you sure that, in such a downside scenario, the fund would recover, and recover quickly enough? The answer to all three may well be yes; but it is important to be clear that that could be the effect of using a long-term risk model. And that would mean – if the recommendations differ – using a long-term model over a short-term model would fundamentally be the more risky option.
In summary then, some processes are so complicated that the only sensible way to model them is with a long-term risk model; DC pensions are probably a good example. Moreover there are many other factors that go into choosing between models, and these may be of far greater significance. However, it is important to realise that there are costs to using a long-term risk model, particularly simplicity and clarity. Moreover, it may actually be riskier than using a simpler, shorter-term model, and expose a fund to taking larger, more concentrated risk positions.