Are LDI models less perfect than they seem?

Rakesh Girdharlal: Most interest rate models, historically, have been flawed, in that most of them assumed interest rates can never be negative. The natural perception is that it has to be asymmetric and it is more likely that interest rates will increase than fall further. I think that is a trap most people have fallen into.

I am not sure trustees will necessarily understand what models go into this asset liability modelling in any detail, and may be misled by the outcomes they see.

Simeon Willis: My experience is that models show a very marginal benefit from hedging at the higher levels, because people often look at value-at-risk as their central risk measure.

This idea that LDI has a diminishing reduction on risk is actually quite misleading; I think it has a linear reduction on risk. For every pound you hedge you remove a pound of interest rate risk.

So when schemes conclude, ‘It is not worth doing the extra 10 per cent because it is a very marginal reduction of risk,’ that is actually a limitation of using VaR as the only risk measure. If you look at it using other measures, such as sensitivity or scenarios, you will see that the risk reduction is much more beneficial.

Barbara Saunders: There are different types of modelling, but if you have an economic scenario-type model, quite often there is an assumption of mean reversion, and you have to set some sort of mean. Therefore, it is very subjective as to what the model produces and may be very skewed by the views of the people who built the models. If they believe a mean is high or low, that will determine whether or not it is a good idea to hedge.

So it has to be built up much more on an economic argument, rather than on a model-based risk argument, and you have to explain the reasons why rates are where they are and in which direction you think they are going to go.

Girdharlal: But if you do believe it is skewed one way or another then that is an active decision you are placing a bet on. And as Simeon said, ultimately, increasing your hedge reduces your risk pretty much linearly, but you have to factor in that the amount of interest rate risk that pension schemes have today is significantly more than the interest rate risk they had at a time when yields were much higher.

It is a very ugly word, it is called convexity: as interest rates have reduced, the sensitivities of liabilities to interest rates have increased.

Giles Payne: The model risk is one thing which, as a trustee, I really do not see discussed enough. We see different consultants come in with different models and what we do not understand is how sensitive those models are to the assumptions that are put in.

Even if you take it as simply as normal distribution – and we know the financial markets are not normal distribution – VaR figures are going to be way out.

There is opacity in the market about those models and how they can really influence the outcome of a decision by just tweaking a few little bits. I think that is dangerous because it is trying to say, ‘This model is perfect’ when it is not, and you should be considering a range of models to see the outcome.

Benoît Jacquemont: One development underway to make this long-term ALM model a little bit more in tune with what the market is doing is that they are trying to describe the world in different time periods, to model short-term regimes varied with a weight towards how markets are behaving right now, and decouple the long-term expectation of return into different time horizons and different multiple regimes.

So typically, the short-term regimes are very skewed towards market behaviour. Then you have a second, slightly longer regime, which is based more on the business and economic cycle. And then finally you have what we call the mean reversion, the long-term trending. These are supposed to capture the real behaviour of expected returns across different multiple periods of the time horizon of investing a bit better.

Pensions Expert: Do trustees accept models too readily?

Payne: I do not think they have the wherewithal to challenge them.

Huw Evans: I do not think they really get any visibility around the modelling. Generally, the consultants will arrive with results based on their model, and they typically are only running one model. So that discussion is not happening with the trustees at present.

Saunders: It is funny though – we have always been quite sceptical of modelling, but we have been pushed to do it because, ‘Actually, our other consultant used to show us this’, or, ‘The regulator has recently come out and said they want VaR in the scheme return’. So rather than being able to build it up from first principles we have been required to provide the modelling, even if we do it with various caveats.

Payne: I think you go back to scenario testing. I personally prefer looking at a portfolio and saying, ‘What sort of scenario is going to break this portfolio, for our purposes?’, and seeing if we can think of one.

Then you say, ‘Actually, that is not a totally stupid scenario, we do not think this portfolio is going to be as robust as we thought it was.’ Whereas, relying on VaR will give you relatives, but I do not think it can be relied on absolutely because it is so sensitive to the inputs.

Jacquemont: I think the key is really to use different risk metrics, as you already mentioned, to assess the risk of the portfolio. Typically you would just think about the volatility of your asset, you think about how volatility is versus equities, you think about specific testing of your investment opportunity across different scenarios in a deterministic way, just by looking empirically at how this kind of investment has been sensitive to different economic market environments.