Quantitative Arbiter (TM) is software designed to keep our clients in finance out of trouble ... not with regulators, but with a cold hard reality that doesn't care whether or not what you've been doing falls within the bounds of "accepted industry practice". In this cold hard reality that we share, all that matters is whether your investments make money or lose money. For clients making investments guided by large-scale risk models, whether or not they turn a profit is driven by whether or not their models are producing good predictions. 

Quantitative Arbiter (TM) tracks how well your risk models are predicting real outcomes. This not only gives you an early warning when things start to go awry; it also allows you to identify and track down smaller problems in your model in the course of normal operation. If you are involved in large-scale model-informed investment, and your organization isn't large enough or well-connected enough to count on a bail-out when things go wrong, then Quantitative Arbiter (TM) is an essential tool for your organization.

Application:

Quantitative Arbiter (TM) is specifically designed to measure predictive error in large-scale, flexible, probabilistic risk models (e.g. mortgage prepayment models, credit risk models). It requires both model predictions and model validation data to run. As input, it takes a large number of individual probabilistic model predictions (e.g. whether or not a certain loan will default in a given period, and if so, how badly) paired with the actual realized outcomes (e.g. the actual payment or default amount for that loan in that period). As output, Quantitative Arbiter (TM) returns a statistically rigorous estimate of a model error metric called the Cumulative Discrepancy. Quantitative Arbiter (TM) has utilities that allow you to then plot or query that estimate to find out how confident you can be that your risk model is (or is not) performing as required.

Before deploying Quantitative Arbiter (TM), it is necessary to first establish what values of the Cumulative Discrepancy constitute acceptable (and unacceptable) predictive performance. Figuring that out only takes a little work, but it does require that small amount of work. Next, your team may require some assistance in developing the connections between model and data that are necessary to produce the input to Quantitative Arbiter (TM). If you decide to lease Quantitative Arbiter (TM), we will assist with both of these tasks to whatever degree is necessary. 

Once Quantitative Arbiter (TM) is running, you will find that it not only provides a hedge against unexpected loss; it also makes your quantitative staff more effective on a day-to-day basis. Right now, validation in finance is done using a combination of visual inspection and misapplied statistical tests. With the clarity that Quantitative Arbiter (TM) provides, you will find that periodic model validation takes much less time overall. And, when you do spot a problem, you can apply Quantitative Arbiter (TM) to sub-sets of the model population as your quants track down the source of discrepancy that is plaguing your model. In short, if applied correctly, Quantitative Arbiter (TM) takes the ambiguity out of finance. If that interests you, reach out to us.