StableSims: Optimizing MakerDAO Liquidations 2.0 Incentives via Agent-Based Modeling

Greetings to the MakerDAO community!

I’d like to introduce myself and @sehyun, we’re both students at UC Berkeley and members of Blockchain at Berkeley, where we co-headed the Education department together.

Since our first semester in the club, we’ve been working on StableSims, a research project in which we created a simulation of the Dai stablecoin system and sought to optimize the means by which the new Liquidations 2.0 incentive parameters, chip and tip, are set.

We’ve finally wrapped up our efforts and have released:

  1. A research paper
  2. Our simulation code
  3. A quick synopsis via a Twitter thread (show us some love <3)

A TL;DR of our findings:

  1. Broadly, our results support the current parameter settings of chip: 0.001, tip: 300. However, there were many assumptions and simplifications made in our research (all of which are listed out in the paper) that make us cautious in claiming that these are the optimal parameters.
  2. Our results show that it is more cost-effective to increase tip as opposed to chip. Namely, when comparing to the performance of what we deemed to be the “minimal” parameter setting of chip: 0.001, tip: 100, the percent decrease in the average time it took for a keeper to liquidate an unsafe vault per additional dollar of incentive spent was 1.5x higher when increasing tip than when increasing chip (while holding all else constant).

The main, high-level takeaway from our research is this second finding. We feel that this also supports relying more heavily on tip in order to incentivize liquidating small vaults, as opposed to simply increasing dust. However, note that the keepers in our simulations did not attempt to execute the incentive farming attack listed out in MIP45.

As a disclaimer: we are not professional researchers. A large part of our goal with this project was simply to learn: to learn about stablecoins, to learn about MakerDAO, to learn how to build a simulation system like this, to learn how to do self-motivated research. So please, take everything with a grain of salt.

And most importantly: please make use of our code! The simulation engine is open-sourced so that you smart folks could take it further and use it in ways that we couldn’t have the foresight to envision. We sincerely hope it becomes a useful tool for the protocol.


How great to see these initiatives of future geniuses.

Do not doubt that possibly in the future, you are the ones who know me about the maker protocol and can be more part and life in the development of the same, with this great contribution and give a great precedent.

Many successes.