Understanding Model Risk

Model Risk shows why we spend all this time using real world numbers to prove or disprove the Model Predictions. In Hedging, all sorts of Model Risk rears its ugly head while trying to predict outcomes, and one must be mindful of it. Option Valuation depends on Volatility, and it is totally unpredictable. We hedge that Volatility, but with Options, so there we are again at Square One, guessing the Volatility of Volatility Itself. It is a Risky Business like dealing with Manufacturing High Explosives, where we are looking for reward from Managing all that Risk successfully, which we are. If we weren't, this would be a funeral for many at once. If she blows, we want to be at a safe location but there will be lots of other casualties. We even hedged that so that there will be someone to organize the funerals, cheaply of course. They all wanna dance on our grave; Make it BYOB.

Look at the underlying. This week, the S+P SPX looks like it was left in the hands of a hyperactive child with a cranky pout on the controls. We get those double digit swings daily. And like a good parent, we ignore them. They'll be gone to Camp in a week after option expiration. They'll behave if they ever want us to pick them up again after their siblings have softened them up a bit.

There is a lot of Execution Pricing Risk. We try to minimize that by making a small amount of trades initially, and later to add and/or adjust the balance. We have also seen that the Model itself may have a bug or two, resulting in a zero value for a parameter that is anything but, and in a crucial location like the Portfolio Risk Analysis Diagram. I wonder how many people in the Detonator Factory are aware of that? It is not like we are taking out Ads; We need someone to counter party this. In a lot of cases, we are the only buyer since everyone else already took off. It goes with the territory; Chicken Dinner.

When it comes to liquidity risk like may be the case on the Nat Gas Hedge possibly born of cranky hedge customers in July, it is hard to quantify by a model. Management would say the simple answer looks good, so go with that. I call it Prozac risk where the happy answer prevails, and it is most likely a bad one. That of course would take care of itself Toot Sweet by destroying the company and management with it. Management then blames you and the model even though forewarned as they scatter to sop up all the available jobs. There is considerable Back Stabber Risk.

Gamma risk is small; 150 K. Vega risk a lot larger; 5 million. Delta downside about 3 million. We get a lot of money by carrying Theta through Thursday, but our risk is maximal on Wednesday, - 18 million At The Present Money, as they will try to shake us I surmise. We get it all back by holding on. But then that is all based on our Model. See what I mean? Things may get sketchy in the multiple unknowns between now and convergence at expiration.

Liquidity Risk: Here are the best of breed ETF's with volume. Many of the regular substandard performers just don't have the volume at times. Volatility via the VIX expires Tuesday as well. That may be the spike we see coming Tuesday-Wednesday in the option metrics.

I do have a unique metric for judging models. Judging the salesman can tell you a lot about the outside of the box model risk. What ever prompted this guy to trudge the streets in the brutal dark cold of January to sell Nat Gas hedging contracts? What is the cut of his jib? Same thing with financial advisers. Nice office. Maybe a little TOO nice if ya know what I mean... Is that a Ming Vase??? Maybe it is just a knockoff made by a guy named Ming on the outskirts of Beijing. It tells you something. Then again I might not care if his Rolex has two "L"s in it if he can bring home the bacon.

So back to the drawing board. Yet we can Test until we are blue in the face if we are not mindful of the Other Model Risks, and we can recognize other previously Unknown Unknowns as they crop up versus our Known Unknowns. Chances are others have been there before, but you can bet they don't exactly want to advertise it or even have the means to from where they landed that 1 time out of 1000.