So I’ve begun collecting data on the behavior of vaults that have been liquidated. I’ve learned a few things, but also have a lot of questions that require further data.
Most users who are liquidated continue to use their vault. This was fairly surprising to me. Across almost all ilks, the strong majority of vaults that suffer a liquidation event continue to be used. The only exception with >2 liquidations was the BAT-A ilk, which has suffered 188 liquidation events with 74 of those vaults continuing to be used.
Use was defined as at least one deposit or generate operation after a liquidation event.
We do not appear to have repeat offenders. Of the total number of liquidation events and the total number of unique vaults that suffered them, there was a surprisingly close match. Only 21 liquidations ever appear to have occurred to vaults that have suffered them before. So that was encouraging.
My conclusions from these two metrics was that we have a surprisingly sticky customer base.
Those vaults that suffer liquidation and continue being used are on average less likely to suffer liquidation events in the future. Note that this is not time-weighted, asset-weighted, or activity-weighted, so it is not necessarily apples to apples, and should not imply any particular policy response.
…which brings me to my request for more data. What I have is aggregated data, so I do not have a time series to compare ilks only in time periods where they were open. While it has been useful for some qualitative data, it needs more depth to answer many questions.
I should note I also have the aggregated data for median and average fees per vault for every ilk. The difference between the two suggests a strongly positive skew – but again, I do not have this data in a time series, so it is hard to compare across different regimes of gas pricing, even within an ilk unless it is a fairly new one. But for most ilks with large numbers of vaults, the average fees earned is several thousand times that of the median.
I hesitate to strongly draw conclusions from that, as I need the time series data for fees if anyone has it or knows how to easily pull it. But it does on the face seem to vindicate a focus on larger vault users across all ilks.
If you have any of the following data already at hand, I would love access to it:
-Average and median fees by urn in a time series like weeks or months
-Average and median CR by urn in a time series like weeks or months
-Length of time from liquidation event until vault closure (or lack thereof) for those that are still active after such an event
-Any other data that is interesting with regards to fees paid over time by various ilks, or about the behavior (both before and after) of vaults that suffer liquidation events vs those that do not
I will end by saying I think this highlights an importance to better familiarize ourselves with which kinds of customers are higher/lower risk and higher/lower profit margin. The median BAT-A vault, for instance, has only generated 0.53 DAI in fees in a median life of 49.5 days. There’s lots of one-off surprises, but it’s hard to ensure they’re not heavily skewed by changes in gas prices. ETH-C in particular stands out as one to search for new customers, as even though it is our newest crypto vault type, the median vault has earned 39.30 DAI in fees over a median life of 4 days.
I think there are probably lots of insights we can gather about the risk profiles, profitability, and use cases of our customers if we can begin to formally build out some data sets on historical behavior of vaults.
Big thanks to @tmierzwa who supplied much of the data I have at the moment. The rest was pulled from Dune Analytics.