Background
As discussed in @monet-supply’s post Adjusting COMP distributed model, it is clear that there is quite a bit of optimization that can be done to the COMP distribution model. We (Gauntlet) believe that the network should begin to increase the incentive for long-term holding of COMP relative to short-term holding and trading by yield farmers. While the introduction of COMP led to the ‘agricultural revolution’ and the usage of liquidity mining as a growth tool, it has become clear from on-chain and off-chain data that holding times, voter participation, and ‘real’ borrowing activity can be improved by distribution tweaks.
By distributing COMP to borrowers, the protocol encourages recursive leverage, which both brings little value and reduces security of the protocol. Based on the survey in that post, it is clear that at the very least, the COMP emission system needs to three control mechanisms:
- Adjusting COMP emission rate
- Scalar multiplier to allow for more uneven distribution between borrowers and suppliers
- Time-locking mechanism
Proposal CP011 demonstrated that a reduction in the emission rate (e.g. via a call to _setCompRate
) can be a useful tool for alleviating both sell pressure on COMP and as a way to encourage long-term COMP holdings.
However, as DeFi has become increasingly interconnected via farming and COMP has been utilized as a farming asset in a number of protocols (e.g. YAM), we believe that these changes should be made in a cautious, security-forward, and incremental manner. For instance, if the COMP emission was reduced 90% and subsequently, a large yield farming incentive led to COMP holders putting large fractions of the supply in a third-party contract, then we can expect both a supply-side crisis and an increase in recursive leverage.
To handle the fact that any change to the COMP distribution will ripple through the ecosystem, we propose a making changes via series of proposals. Each proposal will make an incremental update to the protocol which will be monitored for two weeks before the next incremental state is taken. This way, we can observe the market reaction (both in asset price, volatility, and usage in farming) and use the three control mechanisms to adjust the protocol. This will allow us to utilize an online learning approach to quantitatively choose the parameters for each of these incremental adjustments.
The sequence of incremental proposals that we will use is the following:
- Reduce emissions rate (no technical risk, yields estimates of market demand for COMP)
- Adjust the distribution to suppliers and lenders (minimal technical risk)
- Add in a vesting period / time-lock (larger technical risk, would be the proposal with the largest risk so far)
Parameter Selection and Justification
We have constructed a Monte Carlo model (partially described in a working paper for optimizing emissions that can be found here) that shows that the current market demand for COMP in both centralized and decentralized exchange implies that COMP is being overproduced by 35-40%. However, this can also be illustrated more simply via some descriptive statistics and illustrations. Using tagged exchange addresses provided by Flipside’s Will Price, we looked at the both the distributions of holding times — how long an address held COMP issued via the comptroller before they sent it to a tagged exchange address (CEX and DEX) — as well as the amount of COMP that was held over time.
First, lets look at what fraction of COMP wealth was kept by addresses that didn’t sell all of their holdings (e.g. amount farmed - amount sent to exchanges > 0)
This distribution is unimodal [0] with a median of roughly 56%, implying that 50% of COMP holders kept 44% of their COMP earnings (e.g. used in farming, hodl’d). However, include COMP farmers who sold all COMP they earned we have a much more skewed distribution:
This is strong evidence that large fraction of COMP holders are selling all of their earning and not staying long-term holders. This data shows that roughly 17.2% of COMP issued is completely sold on exchanges and not held by long-term holders.
If we look at the length of time that COMP farmers hold COMP on their balance sheet, we find that most short-sighted COMP farmers sold their assets via DEXs such as Uniswap and Balancer:
(Note that the x-axis is the base-10 logarithm of the holding time)
Given these descriptive statistics and combined with some cursory simulation results, we believe that emissions should be reduced by 20%. Currently, emissions are 0.44 COMP / block and we believe that it should be lowered to 0.352 COMP/block. Once this change is made, Gauntlet will observe the changes to on-chain behavior and market data to estimate whether further reductions are needed.
Proposal Description
We will call _setCompRate
on the uint input 0.176e18 to execute this reduction.
Post-Proposal Analysis
If the proposal passes, we will monitor the following data sources for changes after the proposal is executed:
- COMP trading volumes, order book liquidity, DEX liquidity
- Usage of COMP in other farming protocols (e.g. YAM, YFI, etc.)
- Changes to borrower / supplier statistics (e.g. avg. loan size, avg. borrow)
We will combine these with our simulation model to estimate how much excess supply and/or demand there is in the market post-change and use this to guide our next emission-related proposal.
Proposal Timeline
- Three (3) days for community feedback and analysis on this post
- Proposal submission on August 26, 2020, 3pm UTC / 11am EST / 8am PST (proposal will incorporate feedback)
Acknowledgments
Thanks to Hsien-Tang Kao, Rei Chiang, John Morrow, and Victor Xu from the Gauntlet team for feedback and analysis.
Furthermore, thanks to Flipside Crypto and Will Price for providing tagged exchange data that was used for descriptive analysis and well as for simulation stress testing.
Finally, thanks go out to Peteris Erins of Auditless for feedback and analysis.
[0] The blue line is a plot of the probability density function of the best fit Beta distribution, which is a one-parameter family of unimodal distributions on [0, 1]. There are a number of reasons for comparing this distribution to a Beta distribution, which have to do with the generative processes that one can construct to sample a Beta that can represent the behaviors of the COMP holders in the system.