HackMD
How we did the RAI Experiment
tags: Presentations
Up to date by November 2022
Authors: Danilo Lessa Bernardineli (BlockScience) and Michael Zargham (BlockScience)
Intro
An experiment for validating our theoretical knowledge around an real cyber-physical system: RAI.
The Scientific Method works by theorizing about the system causal inner workings and hypothetizing about the response when performing an intervention. By doing so, an Experiment can be prepared by describing the intervention and data collection methodology, on which the resulting measurements can be interpreted in light of the initial hypothesis.
The subject of study was the RAI SAFE and Controller Dynamics, which we've helped to formalize and tune it before launch. As we'll see, we've sucessfully demonstrated the expected system properties through an single-agent intervention.
Resources
- What is RAI:
- What was the Experiment context:
- What was the RAI Experiment:
- The Experiments results: RAI move simulation (fictious scenario)
Appendix
DT runs
Experiment Results (actual data)
By Full close
- On an 1160 USD / ETH environment:
- ETH hold: 10.8% ROI
- USD hold: 7.4% ROI
By Partial close
ROI
We have closed ~90% of the RAI position while operating on an position-wide ROI of 12.4% over ETH hold. The current situation is described on the simulation spreadsheet.
On an pure USD Hold scenario (all the ETH allocated initially was converted to USD, and we simply did hold it), our current ROI would be 46% rather than 12%
Partial close decision making
One way of reading the situation, is that we did consolidate everything that we allocated to the experiment-rai initiative, and the remaining position is our net profit at any time. The lower boundary for the experiment benchmark is an positive but close to zero ROI over ETH Hold. The average and the upper boundary will depend on the evolution of the remaining position.
We've decided against closing 50% of the position based on an reversibility criteria: we can always open or expand the position, but the window for consolidating gains can be limited. The ETH Long position is far from risk-less, and we should have an clear strategy on how we want to approach it, especially considering that there are some asymmetries in terms of the movement. Without having well-defined defined time-windows and ranges of expected values, there's a lot that can be lost, especially if we react by instinct rather than by deliberation.
Determinant Analysis on the ETH Hold Results
The main take-away is that the leverage on the ETH price did represent around 60% of the net result so far. 22% were due to the initial arbitrage movement, and 18% were due to the redemption price / debt changes. Applying those percentages to the numbers above would allows us to put an exact number for what were the determinants of our net result
The ratio between the RP changes and the increase in RAI debt was 1.9/0.2 = 9.5x. This means that for each 1x RAI increase in debt, the liability value went down 9.5x. This is compatible with an -19% avg yearly redemption rate
