Inside Amonis’ Award-Winning Risk Programme

With new factor-based risk models for both investments and liabilities, Amonis OFP carried off the gong for best Risk Management at the most recent IPE Awards. What sets this Belgian pension fund apart from its peers?

Tom Mergaerts
Chief Executive Office, Amonis OFP

Amonis, a modest medical sector-focused pension plan, and its CEO Tom Mergaerts rarely seek media attention. Yet their innovative approach to risk has drawn the eye of the industry. Taking inspiration from the world of hedge funds, the team has developed a risk factor-based decomposition of the investments and another for the liabilities.

As Mergaerts explains in the Investor Spotlight interview below, the new model is based on a deceptively simple question: “what is the risk that we become underfunded at some point in the future?” Yet, while it is founded on fundamental questions of solvency, the approach rejects the fake precision of Solvency II-style approaches in favour of being “approximately right,” “able to evolve” and focused on “the real risks … rather than hypothetical risks that are latent but not really very important in practice.”

Despite the evident complexity, granularity and ongoing refinements, Mergaerts is determined to keep things simple: “At the end of the day,” he says, “this is supposed to be an instrument for the governance of the fund … it needs to be transparent.”

bfinance: Tom, it’s great to have you with us to talk about how the risk management framework at the Amonis pension scheme has evolved. First, a little background: how did this new model come about?

I suppose you could say that the journey towards our current risk management model began in 2013. That was the year that we introduced an LDI programme in order to hedge out the risk of the very high guaranteed interest rate on our pensions – between 3.75% and 4.75% annually until the expiry date of the pension.

That was quite straightforward, I suppose. But, once that hedge was in place, we decided to take a closer look at the funding ratio. What we really wanted to know was: what is the risk that we become underfunded at some point in the future? We wanted a model that we could use to stress-test the funding ratio – the solvency – of the pension scheme going forward, and for that model to take all the relevant parameters into account.

It was an interesting time to be thinking about our solvency, from the perspective of new regulation and developments in the insurance community. This was when Solvency II was coming into effect and, in our mind, it was a completely wrong-headed approach for long term pensions savings. It’s based on financial mathematics and aims to be very granular (with heavy use of proxies, which are not actually correct); this approach to solvency often ends up being entirely counterproductive. We wanted something more sensitive and sensible, more realistic; we think it is better to be approximately right than exactly wrong.

HEDGE FUND INSPIRATION

Looking back, it would be fair to say that we first got the idea for our factor-based risk model from our experience investing with Key Asset Management [a fund of hedge funds acquired by SEB in 2007]. They used something like this to understand the risks of their hedge fund portfolio. It occurred to us that, in some ways, you can think of a pension fund as being rather like a long/short fund: the investments represent positive cashflows (long), and the liabilities act like a short position on the balance sheet.

What we did, essentially, was to make a risk factor-based decomposition of the investments and also a risk factor-based decomposition of the structure of the liabilities. That model was developed in 2016-17 and we started getting the first results for it two years ago; since then we’ve been using it and building up the time series.

I have to say that the liabilities side of this was more challenging than the – relatively straightforward – investment side. We had to think about how we were going to model liabilities so that they would fit into the risk model. We essentially built a portfolio of zero-coupon bonds and fed it into the model to reflect the rate sensitivity of the liabilities.

bfinance: do you use the same risk factors for the investment side and the liability side?

We actually have two separate models – one for the funding ratio and one for the investments. They’re different because we only want to use the relevant ones for each. For example, the equity risk factor is not very relevant for the funding ratio, yet is significant when we look at the investments only.

It is interesting to observe how the various factors have more or less importance in the different areas. When you look at the investments in isolation, the biggest risk is the market and the second is interest rates. However, if you look at the funding ratio then the interest rate risk almost disappears because your bonds are serving as a hedge for the funding ratio, so other factors become more important in explaining the risks.

bfinance: Since 2017 this model has been in effect and you’ve been collecting data inputs for it. Has there been anything that surprised you? Have you had to make any changes?

Absolutely. A risk model should be able to evolve. You design things as well as you can and you hope the model is well calibrated, but then you see things crop up and you have to explain them, you have to refine things. This is the absolute opposite of a conventional “solvency” approach: there the model tells you what you have to do; here the model is more about learning and understanding the risks. It’s not about being prescriptive. This means that you think a lot more about the real risks that show up in the fund, rather than the hypothetical risks that are latent but not really very important.

For example, one interesting observation – which we have to examine further – is that in the funding ratio we have seen a rather strange volatility element in the risk factor decomposition. To put it simply, volatility seems to be reducing our funding risk rather than adding to it, which is quite counterintuitive. One possible explanation, which I’m currently examining, is that this could be caused by a feature of our pension system: every year the board makes a discretionary decision on whether or not to add an extra profit distribution above the guaranteed interest rate. If you look at those profit distributions in hindsight, they act somewhat like a call option that’s attributed to plan members; perhaps this causes the implied volatility that we’re seeing. It’s something we’re still looking at closely, but the point is that this is quite exciting – we are seeing the model uncover things, we have to try to explain them, understand them.

It’s worth noting how the model performed in late-2018, when there was a significant decline in markets. We have an upper and lower bound on the risk of the funding ratio – which is very interesting to have, actually – and the bounds of risk were not surpassed. It’s quite reassuring that we can measure the risk to the funding ratio in this way.

bfinance: Looking ahead, are there any improvements that you’re considering making?

Earlier I mentioned the liabilities – we use a discount curve (European government bond yields) to make the present value of the liability cashflows. It’s safe, but it’s not optimal. It’s very expensive to hedge a 4.75% return with a bond yielding close to zero. If we could replace that bonds portfolio with a mixed portfolio (e.g. other bonds, absolute return-type investments, maybe mortgages) that could be more efficient.

I have been thinking, therefore, that it might be possible to create a kind of factor-based discount rate. Rather than discounting at the government bond yield curve, we could replace it with a return expectations curve which is created using specific factors. This is certainly challenging: factors are used to describe risk and analyse risk, but forecasting risk is difficult. We will have to do our mandatory ALM study in September, so I will try to examine this further and reach some conclusions before that takes place.

bfinance: Risk factor analysis can be immensely complicated. How are these issues communicated, for example to the board?

At the end of the day, this risk framework is supposed to be an instrument for the governance of the scheme. That’s what it’s for – it’s to help everyone understand our most important risks at the highest level. So the reporting and analysis needs to fulfil that purpose: even though the finer details are of course very important, we need to keep things clear and simple, focusing on high-level conclusions that are easy to monitor.

Once per quarter we present a report to the Asset Allocation Committee and Board. That presentation goes into some details, like the Value-at-Risk, the stress testing, the risk compression ratio [otherwise known as the diversification ratio, this measures how well constituent funds act together to reduce aggregate portfolio risk, i.e. a measure of the efficiency of utilisation of the risk taken.] But it’s important not to make things too cumbersome – it needs to be transparent, although there is some granularity in the report in case the board is interested in looking further.

bfinance: How do you use this risk model when considering changes to asset allocation?

The model has been used several times before a change to asset allocation, in order to understand what the impact might be on overall scheme risk exposures. This provides more comfort before the decision is taken and prevents surprising and undesired risk exposures popping up.

We don’t do this with every asset allocation change. For example we recently decided to add listed infrastructure and didn’t run it through the model in advance because it was essentially a diversification of the equity portfolio and, as we had already observed, changing the composition of the equity portfolio does not really affect the factor exposures much.