The Danger of Feedback Loops

Let’s imagine we have two places to live: a city or rural/suburb. At the outset most people prefer to live to the city because it makes for a shorter and easier trip to work and most of the people they know also live in the city. The suburbs are nice but nonetheless remain an alternative partly because they are not as well developed. Since most of the wealthy people live in the city, conditions tend to be nicer because the higher tax revenues allow for better schools and infrastructure.

But now consider some type of exogenous shock that causes some people to migrate to the city. Perhaps it is because new transportation methods exist that make the once perceived long commute from the suburbs to appear more feasible. Or possibly the emergence of mass production in housing has resulted in the building of brand new neighborhoods that would serve as the perfect opportunity for a couple looking to start a family. Maybe the businesses that were once at the city’s epicenter have been replaced by competitors abroad. In the wake of these changes, the wealthiest people decide to leave the cities for the suburbs.

Now that the wealthiest have decided to leave, the city no longer benefits from high tax revenues; the suburbs on the other gain the advantage. As a result city conditions begin to falter while the suburbs become more attractive. That then leads more people in the city, who may have been the next level of wealthy, to leave for the suburbs. That leads to another round of deteriorating conditions (less tax revenue leads to worsening conditions) as a pattern begins to emerge. But since the least wealthy don’t have the opportunity to move as easily, they become the primary inhabitants of the city; consequently the level of crime grows. That makes the city even less attractive to potential movers. This feedback loop ultimately results in the fall of cities and the rise of suburbs.

This cycle essentially occurred in the United States during the 60s and 70s. This example illustrates the power, and in many ways dangers, of feedback loops. Most of the world is governed by some sort of cycle or pattern, so it is important that we gain a good understanding for the mechanisms for the most prominent cycles.

Feedback loops are a huge factor in financial markets, particularly as it pertains to credit easing and contraction. Let’s consider an economy that is gaining momentum and credit is easily accessible; lenders determine how much they are willing to lend based on the borrowers’ projected cash flows to service the debt, net worth collateral, and their own capacities to lend. If interest rates are low then risky investments may appear attractive, causing those asset prices to go up. But as asset prices increase so do the net worths of individuals, making it appear more likely that those who borrowed will be able to service the debt (since their collateral is a growing investment). That leads to more lending since individuals seem more creditworthy; but that in turn drives asset prices higher as a cycle begins to emerge. Thus credit growth is fueled by itself: easier lending conditions cause asset prices to rise, which makes individuals look more reliable, which leads to more lending.

But what happens once the cycle reverses? When interest rates rise to make debt servicing more difficult, investors may look to liquidate their positions or move into safer assets to ensure they can make their interest payments. But that results in lower asset prices and consequently lower net worth. If lending is based on net worth and collateral then all of the sudden it is less attractive to lend as freely (since you are less confident in the collateral being adequate in the event of a default). As credit contracts, there is less money flowing and lower incomes that further depress asset prices. More selling occurs as the cycle feeds onto itself.

Ultimately the entire economy suffers as unemployment rises and investment portfolios plunge.

These feedback loops, both on the way to prosperity and depression, are what make it a business CYCLE. Optimism feeds on itself to inflate asset prices, but then it’s important to realize when things peak: at what point are conditions ripe for credit to contract and asset prices to go into a tailspin? The tough part is sifting through the false alarms: is this fall a buy the dip opportunity or the start of a marching decline? Answering that question requires careful diligence and a little bit of luck.

Feedback loops also exist in some of the models that govern our lives. In her book Weapons of Math Destruction​, Cathy O’Neil discusses many ways that mathematical algorithms that aim to solve problems efficiently instead create dangerous feedback loops that promote inequality and threaten democracy. Consider the example of predictive policing and recidivism models (recidivism is when a criminal commits more crimes after the first infraction).

Clearly it would be nice if we could figure out who is most likely to wreck havoc after they leave jail and know how to allocate our police officers in a way that best contains crime. So shouldn’t we just devise a model that takes existing inputs and formulates how likely someone is likely to commit a crime, and then adjust the police force to prevent those crimes from occurring? Well not quite.

Consider who is most likely to be deemed most likely to commit a crime: those who are unemployed, live in poor neighborhoods, and know people who have also committed crimes. Some surveys of those convicted ask about first encounters with the police, but given the controversial stop and frisk laws over the last several decades those questions will disproportionately hurt African Americans. If your model puts more policemen in troubled areas then understandably you will catch more criminals, which then puts a greater emphasis on those troubled areas, which leads to more criminals caught. In many cases you are imprisoning family members, causing a destabilization of the neighborhood’s structure.

So rather than efficiently completing the job, these models have developed a self fulfilling prophecy that hurts those in less wealthy areas.

If your model punishes those who are unemployed and live in bad areas worse than others, then once they come out of jail they are even less likely to get a job because they had to serve a more severe sentence. They will essentially be in the same conditions and have an even greater chance of committing another crime, which of course validates the recidivism model! That causes those in poorly developed areas to fall victim to a very problematic cycle. Give worse punishments for those more likely to commit a crime, but that makes them more likely to commit a crime later, which leads to greater scrutiny upon them, which leads to the model validating itself.

So what’s the lesson here? Be aware of feedback loops. Whether it is demographic changes, economic cycles, or mathematical models, feedback loops can have sometimes dangerous consequences. It is important to be well informed and look at a phenomena and ask yourself if there is any sort of reinforcement driving it. Maybe there are feedback loops in your life that have not previously considered, but now can examine in a way that helps you improve as a person.

For example laziness can lead someone to fall out of shape, which makes it harder for them to exercise, which makes them lazier and even more out of shape. Regardless of the outcome, learning about how different mechanisms operate can help us become better students and citizens.

- Bryan Proferes Member (MCG)