10/04/21
The Fed Forgets Lesson #1 of Risk Management; Algorithms and Discrimination: The Need for a Safe Harbor; Complexity, Tight Coupling and Inflation
The Fed Forgets Lesson #1 of Risk Management
Two Federal Reserve Bank Presidents, Kaplan and Rosengren, have now resigned due to questions about their personal trading activities, and how these may have been informed by information that they had by virtue of their positions. Fed Vice Chairman Richard Clarida has also been identified as having executed suspiciously timed trades and may be forced to resign as well (what’s good for the Reserve Banks should be good for the Board of Governors as well).
Risk management is fundamentally about identifying, measuring and managing/mitigating the risks to an organizations objectives. Clearly the single biggest risk the Federal Reserve faces is the risk to its independence, and questions about the integrity of its leaders goes to the heart of this risk.
This, to my mind, is why Compliance either needs to be a part of, or have strong linkages to, a firm’s risk management. At the Federal Reserve, compliance is generally the purview of the Legal department, and here we have a case where an activity may be technically legal, but nevertheless is among the biggest risks faced by the System. My former employer, the NY Fed, has a fairly sizeable Compliance Department, though a review of the Boston and Dallas officers lists do not identify individuals there with these responsibilities (tbf, they may have the role and it just isn’t explicitly disclosed).
The Federal Reserve also has some pretty thorough guidance for banking firms (SR08-8/CA08-11 Compliance Risk Management Programs and Oversight at Large Banking Organizations with Complex Compliance Profiles). Time for the Fed to follow their own guidance.
Algorithms and Discrimination: The Need for a Safe Harbor
A whistleblower, Frances Haugen, who is also a data scientist, has come forward on an interview with 60 Minutes and has alleged that Facebook made algorithmic changes that were designed to maximize engagement at the expense of increasing the sharing of problematic content, such as false and/or divisive information.
"One of the consequences of how Facebook is picking out that content today is that it is optimizing for content that gets engagement, a reaction, but its own research is showing that content that is hateful, that is divisive, that is polarizing, it's easier to inspire people to anger than it is to other emotions," she said. She added that the company recognizes that "if they change the algorithm to be safer, people will spend less time on the site, they'll click on less ads, they'll make less money."
Haugen filed at least eight complaints with the Securities and Exchange Commission alleging that the company is hiding research about its shortcomings from investors and the public.
Facebook of course disputes this assertion
There are a number of aspects we could explore here (Facebook’s incentives; the role of regulation; reputational risk, etc.). I’d like to briefly discuss the effect of algorithms on possible discrimination by financial firms.
Banks and insurers face a myriad of rules that prohibit discrimination on numerous basis, such as race, gender and other standards. Increasingly firms are relying on automated algorithms to make decisions that grant or price access to financial products to their customers, and increasingly these algorithms will use advanced modeling techniques, such as a myriad of machine learning techniques, that are opaque in their decision-making, and potentially biased by hidden correlated variables.
We should expect financial firms to actively avoid discrimination, but as the Facebook example illustrates, looking actively for problems can raise reputational risk (for this reason lawyers will often insist that this type of work be conducted under priviledge). At the extreme, firm’s will avoid looking for discrimination and bias because such evidence can be used against them.
What would be helpful would be a regulatory safe-harbor for firms undertaking a good-faith effort to identify bias and discrimination. Firms should be expected to address any issues identified, and it is probably fair that any specific disparity identified needs to be remediated. But in such cases punitive judgments and fines should be suspended.
Risk management is often about creating the proper incentives: here is a case where regulators can help incent proper behavior.
Complexity, Tight Coupling and Inflation
Charles Perrow wrote an influential book in the operational risk world in 1999 titled Normal Accidents: Living with High Risk Technologies. In the book, he examines a number of infamous ‘accidents’ such as the Bhopal chemical plant disaster and Chernobyl. He observes that there are two principle dimensions of operational risk that one can categorize systems: whether interactions are linear or complex, and whether steps in the process are tightly or loosely coupled.
For the last several decades, companies around the world have been engaged in at least two types of transformations that directly link to this theory. They have increasingly outsourced elements of their supply chains, often overseas, and often to Asia. They have also taken a lesson from the Japanese (principally Toyota) and implemented a just-in-time (JIT) production system.
The advantages of outsourcing were clear; there was a considerable labor arbitrage available as the world globalized following the collapse of the Soviet Union and the admission of China into the World Trade Organization.
The advantages of JIT production was increased flexibility to change products as demand evolved, but a large benefit was the savings from reduced warehouse inventory needs, both in terms of raw materials and work-in-progress inventory.
So using Perrow’s topology, we can see that industrial firms’ production and supply chains have both become more complex, due to outsourcing, and more tightly coupled, due to JIT production. An accident often results from the coincidence of two or several failings, each of which had been anticipated by the designers and operators of the system but which together were totally unexpected. The probability of such a conjunction increases dramatically as the system gets more complex.
Technology has been the primary driver of tight coupling. It has been reported that electronics now account for over 40% of the cost of a new car, up from 18% in 2000 (Deloitte). Chip production itself is an enormously complex process; the largest fabs are in Taiwan and South Korea, but critical manufacturing equipment is made in the US and Europe. Technological progress is generally considered non-linear (exponential) and despite naysayers does not appear to be slowing.
The chip shortage is the result of an accident (covid) affecting aspects of a tightly-coupled supply chain. Manufacturing of chips was affected (Taiwan), key ports were affected (some 90 percent of the world’s electronics go through China’s Yantian port). As PopSci reports:
There aren’t many chip manufacturing plants in the world, and the few that were running during the pandemic were subject to a series of unlucky weather events that delayed the manufacturing process further. Japan’s Renesas plant, which creates almost one-third of the chips used in cars around the world, was severely damaged by a fire, while winter storms in Texas forced some of America’s only chip plants to halt production. Producing these chips also requires a lot of water, and severe drought in Taiwan has also affected production.
So we are left to consider whether this represents a fundamental turn in the inflation outlook, or an exogenously-caused ‘accident’ in a fundamentally deflationary environment. The inflationary impulse here may actually last a few years. With the geopolitical winds, as a country the US will insist manufacturers onshore key aspects of their critical supply chains. We might also see some firms attempting to vertically integrate key resources (this applies to many aspects beyond chips, such as rare-earth minerals for batteries).
Spurious Statistics
Errata
What I am reading: The Global Financial Cycle (CEPR)
What I am listening to: Khruangbin (youtube link)