One Day Technology Will Know What We Want.

This is what experts think what technology will do for or us ... one day, in the near future. One day everything will be custom ...

We have great mathematical schemes that solve a variety of models for financial instruments from the Vanilla to the most sophisticated deal types of equity, interest rates, FX, inflation, commodities and credit - accurately and robust. And we have internally organized them orthogonally. And we know how to perform cross model validation, precision testing and what have you from our machine learning work where automated model selection is indispensable.
And such decisions are made on a huge amount of information that is dynamically divided into training and test data samples.

It was my "hobby horse" to push this approach for UnRisk. We called the project internally UnRisk DIRECTOR. 2005, I remember. We were proud that our engines are fast enough to do all the "across all" valuations. We proudly announced the idea ... but I was wrong.

What I suppressed was the explorative learning effect. Getting insight from failure. Our users get insight from understanding the difference of, say, instrument-model-method triples. We automate across-models scenario runs, as many other work flows, but do not select models automatically. We even do not automatically select numerical parameters of our distinguished solvers.

We stopped the DIRECTOR project years ago. And we know from our users who analyze model dependencies that they manage model risk by insight. With conclusions like reducing complexity and managing the revival of models, like the normally distributed short rate models, in market regimes where interest rates can become negative.

Yes, technologies will have capabilities to become totally invisible, know what we want and decide for us, but, no, in many solutions they should not just "quietly" select methodologies, algorithms or decision rules. They should be white boxes giving options and provide metrics for comparable results.

For explorative, constructive learning - as Aaron Brown points out in his great book: Red-Blooded Risk. Great Risk Managers learn .... remember VaR in the Jungle ...

Technology will become so sophisticated that we won't be even aware interacting with it, but IMO human-machine interaction in sophisticated working environments should have higher goals.