Weekly (almost) Narrative on MakerDAO – 29 May 2019


In the last two weeks, the community voted on and attempted to implement a reduction in the Stability. After almost two weeks of somewhat stalled voting, yesterday the Stability Fee was decreased and is now 17.5% per annum. The community has maintained the polling frequency for another executive vote. At present and pending an executive vote, the Stability Fee looks to be remaining the same. The overall price* of DAI has significantly rallied to its soft peg target of 1.0000 and for the most part stuck to the peg. During the last two weeks, the total outstanding DAI has now contracted to just slightly above 80.3mm DAI. This continued contraction is largely expected to the attributed to the elevated Stability Fee.

The above being said, as the DAI price* continues to hover above the target of 1.0000, we can draw an initial conclusion that most of the market maker inventory has been cleared out; therefore, it is reasonable to believe the supply overhang (e.g. difference between Demand and Supply) has ceased growing and is basically zero. As the supply overhang has disappeared, a new risk appears that might cause the price of DAI to continue to elevate above 1.0000 (even in the face of a recent Stability Fee decrease). As now, the expected primary sellers of DAI at 1.0000, the market makers (which previously  had outlined their desire to sell at 1.0000) no longer have excess inventory. As such, the remaining sellers will not have large inventory to sell which could cause a squeeze on the price upward. This further implies that historic excess inventory was suppressing the price at or slightly below 1.0000.. In the absence of that excess inventory, “true” supply and demand market forces are now at work for price discovery.

In parallel to the above, the underlying collateral for SCD (ETH) has surged in value. As such, if the Stability Fee were to be incorrectly positioned, we should see either the DAI outstanding surge upward or downward. As evidenced in the absence of a large change in DAI outstanding, we can deduce the Stability Fee is more or less priced correctly on a blended-average across all CDP holders.

Further, the average daily maker burned (as calculated) is now right over 50 MKR per day, down from over 60 after the recent Stability Fee decrease. The total MKR in the “burner wallet” has now surpassed 1522 MKR.


With the upcoming introduction of the DAI Savings Rate, we need to start polling for / forecasting where to start the DAI Savings Rate. As it is strongly recommended to treat the introduction no different than another other new market force, it is strongly advised to roll-out the DAI Savings Rate slowing starting at 100bps.


It is expected that the DAI Savings Rate contract will be an essential metric to gauge and estimate demand changes. As such, it is essential for future PID algorithm optimization to gather as much demand side information as possible, hence the logic to start slow and iteratively determine the market correlation and sensitivity.

In a continuation of the general theme, it is recommended to establish a workgroup to start the logic on a Machine Learning PID Control System** tool to help with signaling.  As outlined prior, it can be reasonably forecasted that the system will shift to see a control system move from signaling humans that vote to one where the ML PID tool would recommend tho outputs to implement and have humans vote on the tuning variables (ML PID algo) or to veto all-together.

The PID should start as purely an informational tool (off-chain but pulling chain data among others for computation) for the governance process as a whole to allow humans to vote based on the output. Thereafter, in contrast to the last narrative, it is recommended to use resources to optimize that algorithm to the point where oracles could be used to feed the blockchain (and thereby the smart contracts) the variables in a manner that is resistant to attacks (e.g. the ultimate ML PID should not be on chain). Thereafter, the governance team and community as a whole should then be voting on the tuning of the algorithm while always retaining the ability to veto the output.

Machine Learning mixed PID has made huge strides forward in the last ten years. Today, cars on “auto-pilot” mode can use in-car image processing to detect its surroundings and navigate accordingly. Most specifically, the car maintains itself in its lane (even in a cross-wind or rainy conditions). The algorithm compensates the turning to smooth out external forces that cause the car to drift away from the center or simply disregards the “noise” of the rain. All of that in-car processing still requires substantial computational processing power (especially when compared to the gas required to execute the equivalent on-chain in a smart contract). Interestingly the data that is needed to start a rough version of that ML PID for Maker is virtually all publically available and accessible. Clearly the price of the collateral and other market related metrics are important data feeds. However, why limit ourselves there? If some centralized crypto exchanges would be willing to share their FIAT on-ramp dollar amounts in a zero-knowledge way, we could use those data points to identify the “storm that will cause the wind”. As crypto grows, other data points of investment trends in general start to become more relevant. As complex as we could make it, the net differential impact will be limited as Maker and the community as a whole are able to nimble enough to implement a change to the Stability Fee in a week which is of course substantially faster than any analog counterpart (e.g. making the ML PID “good” may be “good enough” as it will be so much faster than its analog counterpart that the differential benefit for including third- fourth- or fifth- level data on top will not have the same incremental marginal benefit as getting the core to work).

Therefore, if not already done, it is recommended to put a team together (probably needs to be grant sponsored) to spear-head this endeavour. Of course, in this space, the well-placed paranoia about how a system can be gamed remains. For the team that assembles the  algorithm, they will be privy to “almost” inside information. Therefore any such actions / recommendation must either be done with extreme privacy / secrecy or complete transparency.

Capturing and feeding in such data and then determining the best DAI Savings Rate and Stability Fee overall becomes a perpetual iterative feedback control engineering challenge. There will be input variables that we cannot conceive right now. Removing the human emotion side of how to keep the Stability Fee low (the parallel of keeping the car in the lane) is important to long-term acceptance and price stability of DAI as we navigate our way to general acceptance and hopefully not hit too many black swans along the way.

* – price being determined by USD fiat offramp via USDC – DAI (at pro.coinbase.com)

** – PID (proportional / integral / derivative) Control System. It is expected that the DAI Savings Rate will be a potent (and more so than the Stability Fee impact on Supply) tool, as such it will be supply leaning (much like an autonomous car that has an alignment issue and drifts to the right, the control system for Maker will unlikely be symmetric between supply and demand).

NOTE: Not a part of the Maker foundation, just my $0.02 and not intended as advice in any capacity.  

Top 250 MKR holders = 686174.698

1d 🔺: 615.468

1wk 🔺: 1720.449

Live STBLTY Fee: P/E (dilut.) 52.08 – P/E (w/o dev. fund) 38.49

FCST 50bps STBLTY Fee (VaR MKR burn portion): P/E (dilut.) 1833.51 – P/E (w/o dev. fund) 1358.15