December 30, 2017

Current risk weighted capital requirements for banks are a stand out example of “garbage in garbage out”

Sir, when discussing artificial intelligence and “how much power should be ceded to the machines” you mention: First. “the need to overcome limitations in machine learning techniques”; Second. “garbage in, garbage out…the need for better quality control”; and Third. “the need to develop a clear and transparent governance structure for AI”, “The paradox in ceding powers of decision to AI” December 30.

Sir, human intelligence is quite often in need of all that too.

When bank regulators used intrinsic risks of bank assets as inputs for developing their risk weighted capital requirement, they could not produce anything but garbage out. What they should have used is unexpected events or the risk those assets could pose to our bank system, namely the risk that bankers would not be able to adequately manage perceived risks.

And little evidences the need for a transparent governance structure for human intelligence too, as current regulators refusal to answer the very basic questions: “Why do you require banks to hold more capital against assets made innocous by being perceived as risky than against assets becoming dangerous by being perceived as safe?”.

Humans must also also overcome some technical limitations: An Explanatory Note by the Basel Committee on the Basel II IRB (internal ratings-based) Risk Weight Functions” expresses: “The model [is] portfolio invariant and so the capital required for any given loan does only depend on the risk of that loan and must not depend on the portfolio it is added to.”

And the explicit reason for that mindboggling simplification is: “This characteristic has been deemed vital in order to make the new IRB framework applicable to a wider range of countries and institutions. Taking into account the actual portfolio composition when determining capital for each loan - as is done in more advanced credit portfolio models - would have been a too complex task for most banks and supervisors alike.”

Sir, finally, I would add a fourth requirement, namely to make sure artificial intelligence is kept free from that excessive hubris and besserwisserism that too often affect humans. Like that which kept regulators from even having to define the purpose or banks before regulating these,