Olivier

I am humbled by the support for my money market interest rate strategy, but to be honest, it was a proof of concept. The real value of the controller is in CDP rate management. I reviewed the leading CDP infrastructures and assessed their limitations and designed a system that I believe mitigates observed risks with the highest ROI trade-offs. The result is a fully autonomous stablecoin interest rate system that adapts in real time to optimize borrow rates without inhibiting scaling and while addressing long tail risks. The strategy can be tuned for any collateral type, and the phi-strategy concept I introduce can be developed to remove the need to tune altogether. Of course, I didn't give it all away in my research. We are developing this infrastructure at @Cod3xOrg for future implementation in our native credit layers. My latest paper is available here (note SSRN review in progress): Autonomous Money Supply Strategy Utilizing Control Theory @SSRN papers.ssrn.com/sol3/papers.…

Happy to announce my interest rate strategy research paper is now publicly available. I reviewed the leading lenders' strategies for inefficiencies and developed a strategy that mitigates identified risks. The strategy will be live sooner than you think! Adaptive Money Market Interest Rate Strategy Utilizing Control Theory @SSRN papers.ssrn.com/sol3/papers.…

Aurelius. Live. Now.

@letsgethai is an interesting protocol coming to Optimism that uses a PID controller to manage redemption, thus stabilizing debt value. This made me curious as I recently developed my own model for a PID controller that manages interest rates to stabilize debt value, that you can read about here: xitter.gq/crypto_yuvi/status/174… I looked through $RAI architecture (predecessor to $HAI) and $HAI docs and wanted to document some of my thoughts: Note that while I have some experience with control theory, I am by no means an expert. These observations are points of curiosity and I am happy to be corrected. I will continue to research and follow along to see how it plays out. First, they have opted to use a PI controller, making the assessment that D control might lead to too much forecasting risk. Correctly tuned Kd typically stabilizes systems more than anything, so this is a weird design choice for me and I don't fully understand it. You can suppress Kd while you tune Kp and Ki, and Kd can be super empowering for long term stability as the system scales. I noticed that they calculate a pTerm and iTerm and I'm not sure where the derivation comes from because I expected to see just see an error parameter (which would be the pTerm in this case). I haven't seen a separate iTerm calculation before so that is something I would like to learn more about. It's also interesting that they implement a leaky Integral to mitigate the integral term overpowering the proportional term. Typically an integral correction is designed to overpower the proportional gain when error is low, by design. I'm not aware of any control systems that favor proportional gain all the way to settling. I've seen leaky integrals in hydraulic applications but never for this reason, so that's cool if it works as intended. Structurally it is peculiar that they opted to use the redemption rate as the feedback signal when most CDP protocols have identified that redemption is a big UX drawback. Should I ever reach a point of integrating my control system into a debt protocol, I would prefer to focus on interest rates as it targets positions agnostically and favors liquidation, which is much more user friendly than redemptions in my opinion. A key component of my model is a basic market response, which forms its own feedback loop in the feedback path of the main loop. I think it is important because it affects the sensitivity of system response. Tuning 'in a vacuum' has merits for understanding fundamentals but in an economic context the market response rules all. I don't know if or how that has been accounted for here but I suspect some of the threat vectors explored in RAI's tune would be greatly mitigated by second order effects. Anyway I think control systems will unlock huge efficiencies in the way the entire industry handles debt so I am excited to see how $HAI goes. I have tremendous respect for their team for tackling such a challenging problem and I hope to revisit this post many moons from now as a more educated man.

gm, new medium article medium.com/@zeroalpha/quanti… this one is fairly technical essentially, you can pessimistically forecast liquidation unprofitability by compensating slippage by the unfavorable deviation of the liquidation return either slippage is too bad, or slippage is trending towards being too bad either way, you can catch it before it becomes a problem s/o @GranaryFinance for providing data, discussion, and support 💪

Gm brypto nerds, Ever wondered what quantitative finance is and how it works? Well strap in, because if this knowledge bomb doesn’t get you a date then nothing will. Today we will discuss the basics of time series analysis. This is the crux of all quant work and ranges in complexity from drawing lines on a chart, to requiring significant computation. Fair warning: there will be TA jokes. To be clear, if TA helps you make decisions, then you do you. I like to joke about it. If it hurts your feelings, then maybe your TA sucks. “Time series analysis is a statistical method used to analyze and interpret data points collected over time”. Like price charts. It is used to identify patterns, trends, and underlying structures within data, studying temporal dependencies to understand how past observations can influence future outcomes. Like TA. There are 5 key components of time series analysis: 1. Trend Analysis: Examining the overall direction of the data to identify long-term patterns, such as increasing or decreasing trends. Like the silly little triangle wedge things you draw on random price charts. 2. Seasonal Analysis: Identifying recurring patterns or cycles that repeat over a fixed period, often related to seasons, months, or days. Like how the market always dumps in September, probably just because everyone thinks it will, so it does. 3. Cyclical Analysis: Identifying patterns that occur at irregular intervals, indicating longer-term economic or business cycles. Like the 4 year BTC cycle that will continue ad infinitum. 4. Noise or Random Fluctuations: Accounting for irregular and unpredictable variations in the data. Like the blatant market manipulation ruining your shorts. 5. Forecasting: Using historical data to make predictions about future values, helping in decision-making and planning. This is actually fine. Beyond the basic moving average, the most common technique is called autoregressive integrated moving average (ARIMA). To break it down, ‘Autoregressive’ refers to modelling the relationship between the observation and several lagged observations, essentially capturing the temporal correlation of the data. ‘Integrated’ refers to making the data ‘stationary’ by way of differencing, giving constant mean and variance over time. You could present price data as +/- change, or % change. The order of differencing changes based on the data and scope of model, and can get hectic. ‘Moving Average’ represents the residual error of the observation from an aggregation of previous observations. Financial ARIMA models are often computationally expensive and proprietary, especially if they provide a trading edge. They can also be super useful for analytical purposes like performance monitoring and yield optimization. For example, the @Reaper_Farm team employs a simple ARIMA script to monitor (and predict) the performance of strategies employed by the multi-strategy vaults. This allows for more effective allocations, giving the vaults that little bit of extra yield. The key takeaway is to think about the correlation of the variables you are modelling, and how sensitive the dependency really is. Strategy performance over time is mostly influenced by TVL, abstracting many extraneous variables by compounding into underlying assets. On the other hand, the price of your 100x gem over time is probably influenced by more than you are modelling on TradingView, so whatever lag is baked into your model probably makes it irrelevant. Sorry. Another takeaway is to consider the timeframe you are modelling. Using one year of data to predict APR over the next few hours between harvests is massive overkill. You can probably use a 7 day moving average window over thirty days of data and get a reasonable forecast. Similarly, using one week of price action to inform an entry for a long-term investment makes very little sense. What data ranges are you using, and how far are you trying to forecast? There are many questions to ponder. I hope this has been an enlightening read. With enough feedback and direction, I will gladly delve more into concepts, examples, whatever you guys and gals want to know. Let me know what your favorite indicator is in the comments below. Bonus points if you provide some analysis of why it is useful and how you use it to make a decision. I am always eager to learn from those ‘trying things in the arena’, as it were.

The question is not whether a protocol is rehypothecating your assets or not. The question is how. Every dollar you put in your bank account is rehypothecated by the bank. Banks invest, they hand out loans, and so on. While you earn 5% in a high interest savings account that requires you to top up each month, the bank is lending your money to a home buyer at 8%+ and keeping the spread. FTX 'rehypothecated' user funds, and did so very poorly. We know how that ended. You should not avoid rehypothecation. In an entropic market, it is unavoidable. You should seek to understand it. Ethos Reserve rehypothecates deposited collateral. This is not a bad thing. Ethos Reserve rehypothecates collateral to private vaults that predominantly do one thing: supply assets to lending markets to earn interest. Every time you use a lending protocol, you deposit assets and borrow assets wherever interest rates are the best. That makes sense. With Ethos, you borrow stables with zero % interest rate (0.5% issuance fee), while your assets are rehypothecated to lending markets to earn yield. Your assets go where the yield is best. The spread is not kept by the Reserve. It is returned to users as real yield to ERN stakers. That is the best lend/borrow relationship you can possibly find. This is all managed automatically by the rehypothecation engine. Ethos developers can suggest changes to strategy allocations if needed, but they cannot access user funds. This is literally as good and safe as rehypothecation gets, and more importantly, it is sustainable and can scale. Don't be afraid of rehypothecation. Understand it. @EthosReserve