Model Risk and upfront greed

The recent discussions around black swans and hand-wringing around Type 3 risks seems to misplace the observation that models on their own do not lose money, only people's application manage to do that. The classic example is losses deriving from synthetic tranches referencing sub-prime credits where people were booking deals using models that were a shoe-in from the corporate Gaussian model - or in many cases booked as single name CDS - losses attributed in this activity can hardly be blamed on the model only that the traders used and their support staff allowed an inappropriate use.

Models are by definition a simplification of the real world, the choice of model is a trade-off between sophistication and computational speed. Its key role is to value something that is bespoke based on what is available in the open traded market - so that the bespoke premium can be extracted and the risk can be hedged using the same tradeable instruments.

All well and good. The key trick is actually in the risk aggregation process. For example we hedge a variance contract on the SP500 using a portfolio of traded options across many maturities and strikes. The residual risk should therefore be low and hence so should the daily PL volatility so long as the forward variance of the contract is derived from the volatility surface backed out from the listed options. In this example it is standard practice to aggregate the vega exposure of both the contract and the options and declare that for changes in implied volatility the resulting portfolio has low exposure. Since volatility is an input to VaR calculations, the resulting capital charge is also small. The question for a risk manager is whether they can allow the trader to continue building up a variance-option basis - how much vega basis is simply too much and why and should other measures such as stress testing be changed to capture the split between the options and the variance during market disruptions. The difficulty for a risk manager is that they need to be in the detail to spot the behaviour pattern and raise the issue.

Coming back to model risk - both the variance contract and options are based on Black-Scholes where the distribution of market moves has been modified via the spot and forward volatility curves. The inadequacies of the model - in particular to undervalue extreme market events is well understood by the trading community and their behaviour at a micro level adjusts accordingly.

The problem for banks is when as in the structured credit world, the trading community do not fully understand the model issues and have yet to alter their behaviour. This is the result of rapid expansion where all the trading desks are grappling with are trade capture and operational integrity rather than risk management. The movements in credit and their traded tranches should now force them to reasses their risk.

The problem therefore is how the bank allows large build up of basis risk between one strategy that clients are fond of (IR turbos, long equity correlation, long mezz credit, short variance) and the available hedges. Aggregation on which limits are set allows perfect netting that in some circumstances is optimistic at best. It is always important to question excessive margins on transactions and get the finance people to hold some of the money back as adjustments.

At the end of the day, models never capture the greed to book all margins upfront.