When repeated games fail

This has been in my drafts for a while, but Nicky Case’s wonderful game on The Evolution of Trust inspired me to finish it. I’ll cover some basics on game theory and repeated games below, but he’s a far better teacher than I am - I strongly recommend going over there, even if you already know this stuff.

Being a part of the migrant working class in 2020, renting is my default. I have changed places quite a few times now, and the experience has been entirely professional - if expensive - and perfunctory, except for one part: the return of the security deposit. This has never been an easy experience, and I wonder why - and how we can do better.

Now, there is always the possibility that this is an experience limited to me, and caused by me, but I suspect not. I have had the same experience across countries, types of living situations, having done it while single and also while living with someone, and most recently while helping my family move after the passing of my dad. In almost all cases, it did not seem like I was renting from individuals that fell outside the norm, nor did situations arise that felt outside the commonplace. How then does this happen? Let’s consider this from the perspective of game theory.

Game Theory in the real world

Simply put, Game theory is a way of modeling human interactions and their outcomes, and trying to understand how rational agents act when presented with incentives for cooperation and betrayal. The history of game theory has been fraught with problems, with dedicated studies showing that we are anything but rational in our decisions, and that real-life presents us with expected values and distributions of success rather than concrete positive numbers we can bank on. Most of our decision making is lazy, argues Kahneman.

That said, another one of the shortcomings of the human brain is that it’s hard to theorize about complex things, especially when they become recursive. It’s hard for us to hold on to the complexities surrounding individual decisions, and far less so when we add additional individuals, not to mention the infinite chaos of society. The three-body problem might be a good analogy, even though it looks like we might be making some progress on that front.

Game theory is a useful - but not exhaustive - tool to model interactions in a simple way, and to reason about them to understand why they sometimes succeed and fail. The most useful method I have found in recent times is the repeated game.

Again, I recommend Nicky’s example, but in a few words, a repeated game is where an interaction between agents is repeated multiple times, and the agents retain the memory of what happened in previous games. This is sometimes a good way to model how we take into account present gain vs future consequences, and do not seek to maximise immediate gain when it can hurt long-term success.

A famous example is the marshmallow experiment, but a simpler one would be the process of going to work. It may be possible to maximise your gain for the day by stealing from the workplace or embezzling, but this short-term gain is often outweighed by the lost income from losing one’s job.

Repeated games are how our society enforces long-term commitments and cooperation, sometimes loosely through reputation and punitive punishments, and sometimes strongly through contracts that cycle, like rent and employment agreements.

When repeated games fail

Renting is a good example - and the benefits promised by a repeated game generally hold true for the duration of the stay, in my experience. I can maximise my gain by not paying rent, but I lose the security of long-term accommodation if I do so. The landlord can raise the rent on me, but in an even market he risks losing a tenant if he does so. Now there are power imbalances in most markets, but let’s ignore them for this discussion. In this case, the good behaviour of both parties is enforced through an expectation that the game continues.

Understandably, this falls apart when the agreement inevitably comes to an end - which is where we deal with the deposit. In a highly populated city, most landlords and tenants have little expectation that they will meet again or do business together. This makes the landlords’ position quite different - they can try to maximise the amount they retain from the deposit through any means, and have little incentive to do otherwise. Quite an interesting problem indeed.

The same problem plagues the end of most agreements, from bosses being asked to give good recommendations to employees that leave, to VCs behaving badly at the end or failure of an investment deal. This is a rather new problem, posed by the rapid population growth and the change in our interactions caused by population-dense dwellings and places of work.

I spent part of my life in a small village in India with a population under 2000 even at the busiest of times. Everyone knew everyone, and this made practically every game repeated. If you were a carpenter - as my grandfather was - you couldn’t be sure that you were seeing any of your customers for the last time. Even if you were, you can expect to be dealing with their kids and relatives for a long time, all of whom knew you. Communities reduce short-term thinking in this way, and it’s something we’ve lost.

Most review and critique platforms out there exist to inject this back in, by trying to make sure that disgruntled agents in an interaction can try and connect present bad behavior to future bad outcomes through bad reviews. This is a technique that has many downsides - it can be easily abused, for one, or gamed - but it’s what we have.

Another attempt is in trying to reduce how many opportunities exist for bad behavior. We do this through tighter contracts and stronger agreements that try to exhaustively stipulate what can happen, but the complexity of human interactions and the difficulty of challenging something through the judicial system still leave a pretty wide margin to be exploited.

That is not to say that the community-oriented way of solving this has completely failed. Modern solutions like Y Combinator have been - in my experience - very successful in enforcing behaviour through a strong community and links of communication, and increasing the size of a collective agent with whom a repeated game is inevitable. A VC can be somewhat sure that he will never see a particular company again, but the chance that he will have to interact with YC as a super-fund is not insignificant. Very effective, but I don’t think we’ve still found many ways to expand this benefit outside a manageably small community of people.

What then, is our solution? I admit, I do not know. One solution might be to reduce the security deposit as the final months of a contract approach, with the not infallible assumption that the possibility for damage reduces as you get to the end. This might risk shifting the same misincentive in the opposite direction towards the tenant, but short of eliminating it I think the best we can do is to balance it out. Escrow is sometimes a good solution, but isn’t always possible to do in a perfectly trusted manner.

All of this leads me to the larger hypothesis that the now omnipresent community of global citizens (for lack of a better phrase) or third and fourth culture individuals will eventually need to find ways to aggregate and establish collective bargaining. The reason this hasn’t happened yet is perhaps that existing cultural divides are too strong, combined with the fact that this is by definition a distributed group.

I’m still not sure though - perhaps a better solution exists, and as of yet eludes us.

Update: I’m told there are renter’s unions in Germany that take a small fee to help with these exact problems. This makes me happy. Perhaps there is a darker side to it, but for now what little I know makes me happy.