Machine learning, post-Brexit novation and a world after Libor

The week on, March 10-16 2018

7 days montage160318

Quants warn over flaws in machine-learning predictions

Six quants debate whether the tool can adjust to paradigm shifts in financial markets

Brexit: banks take the ‘no’ out of novations

Swaps clauses stop end-users blocking counterparty switch, making it easier to move trades to EU affiliates

After Libor: Japan, Australia look to multi-rate future

Using new risk-free rates alongside Libor equivalents gains industry support


COMMENTARY: Don’t rage against the machine

As the machine-learning boom continues, several quants have begun to warn of its flaws and hazards. They are not wrong to do so. As they point out, the training data set is limited (there’s only one set of historical market prices), and basing forecasts entirely on past data leaves you vulnerable to a sudden shift in market behaviour.

In a time when machine learning is being leant on to resolve issues such as model risk, as well as provide new insights into short-term stock price changes, it is of course worth delivering an occasional note of caution.

But it’s also worth asking: what makes us think humans can do any better? Humans also have only one set of data on historical market prices to learn from, and tend to privilege recent experience, or, worse, data that backs up their existing assumptions. And if critics emphasise the value of human intuition over machine learning, consider this: what have those intuition-based assumptions done for us lately?

The stock price study in question found that one widely held assumption – the belief that intraday pricing and risk models should be sector-specific – was simply not supported by the data. There is an entire subgenre of literature (with Daniel Kahneman’s excellent Thinking, Fast and Slow at its head) that lists and analyses the countless ways in which human intuition can be led astray.

The kind of casual, rapid heuristics that served humans well for most of history simply aren’t fit to handle the massive and complex data landscapes of modern financial markets unaided. In the financial crisis, 10 years ago we saw a very good example of this. Yes, machines are not good at dealing with sudden step changes in market dynamics – but neither are people. Machines, at least, are steadily getting better at it.



The biggest operational risk loss in February was incurred by US insurer MetLife, which reinstated $510 million of pension reserves it had previously released in the 1990s. The firm discovered it had failed to make appropriate efforts to locate almost 13,500 people before declaring them “unresponsive and missing” and releasing their funds from its reserves.



“There is a natural bias toward optimism. Most firms that are in the doldrums will have a route map out of that and will convince themselves and their auditors that a downward trend isn’t going to continue. It would be surprising if someone said ‘my view of the future is calamitous, and therefore I’m going to voluntarily have significantly higher provisions than the neutral projections would indicate’” – Adrian Docherty, BNP Paribas

  • LinkedIn  
  • Save this article
  • Print this page  

Only users who have a paid subscription or are part of a corporate subscription are able to print or copy content.

To access these options, along with all other subscription benefits, please contact [email protected] or view our subscription options here:

You are currently unable to copy this content. Please contact [email protected] to find out more.

You need to sign in to use this feature. If you don’t have a account, please register for a trial.

Sign in
You are currently on corporate access.

To use this feature you will need an individual account. If you have one already please sign in.

Sign in.

Alternatively you can request an individual account here: