DAOrayaki DAO Research Bonus Pool:

Funding address: 0xCd7da526f5C943126fa9E6f63b7774fA89E88d71

Voting progress: DAO Committee 3/7 passed

Total bounty: 100USDC

secondary title

DAOrayaki DAO Research Bonus Pool:

Funding address: 0xCd7da526f5C943126fa9E6f63b7774fA89E88d71

Voting progress: DAO Committee 3/7 passed

Total bounty: 100USDC

Original Author: BlockScience

Contributors: Natalie, DAOctor @DAOrayaki

Original: Introducing Automated Regression Markets (ARMs): A New Price Discovery Mechanism for Semi-Fungible Assets

BlockScience is researching and developing a novel price discovery mechanism for high-dimensional, semi-fungible assets. This mechanism, known as the Automatic Regressive Market (ARM), has multiple potential applications in various markets; our first exploration will delve into use cases for sustainable value energy markets such as Renewable Energy Credits (RECs) and Carbon offset/removal credit (CORC) market.

In collaboration with DLT industry leaders at Hedera Hashgraph and the HBAR Foundation, we are working on a research project to model and simulate the Energy Credit ARM. We will examine the price discovery potential of junior market makers, and the supply and demand behavior of assets with multiple attributes. This is the first in a series of articles on this joint R&D effort.

secondary title

In our paper "A Practical Theory of Fungibility", we introduced the concept of automatic regressive markets (ARM). The idea of ​​ARM is to leverage the mathematical similarity between machine learning (ML) models and automated market makers (AMMs) to provide dynamic price discovery for non-commodity assets. Goods are a set of objects such that the properties that affect their assessed value and demand are equivalent. In short, goods are interchangeable. This concept is at the heart of how existing AMMs work - pooling fungible assets and discovering prices based on their algorithmic curves. ¹

The ARM expands beyond the AMM, allowing semi-fungible assets such as energy credits or carbon offsets/removals with different attributes to be bought and sold on the main market ARM. This provides efficient price discovery for highly complex arrangements of attributes, such as (for energy credits) energy production type, geographic location, etc.

One of the most important observations from this work is that the substitutability of two items depends on the attributes of these items provided and the context in which they are evaluated. This observation helps us better understand supply and demand in markets consisting of non-commodity (or "partially fungible", see below) commodities.

Item attributes characterize the supplier, and items are said to be distinguishable if they have different attributes. The context in which project attributes are evaluated characterizes the demand side. Goods are said to be substitutable in a particular demand environment if differences in the attributes of goods do not affect their evaluations and thus the demand for those goods. The substitutability of some substitutable commodities varies with the demand environment. For example, if a business wanted to buy solar credits produced in the US Pacific Northwest, the buyer wanted assets that would not be a substitute for credits produced in Asia. The ability to identify assets by attributes such as production location or type of production is not only necessary for many buyers, but can have a huge impact on the valuation of those assets.

The plot represents the scatterplot and histogram of the distribution of the first two principal components of the simulated attributes. The auto-regression marketplace utilizes adaptive subspace detection and recursive regression to dynamically (re)discover which combinations of attributes are valuable.

Existing Cases for Semi-Homogeneous Price Discovery

At first glance, the automation of markets that can buy and sell partially fungible non-commodities may seem far-fetched. However, in both industry and academia2, the convergence between markets and ML is prioritized. The housing market fits relatively naturally into the category of semi-homogeneous goods. Homes are all distinguishable, but when their attributes are similar enough, they may be considered fungible (before buying - ownership affects fungibility). For these reasons, online real estate marketplace provider Zillow is likely to increasingly rely on its AI application "Zestimate" to drive buying decisions.

Zestimate may be one of the first examples of hybrid intelligence that can both learn from and create campaigns, but it's important to observe the nature of the model's execution. Zillow has started offering home quotes based on its Zestimate, which will naturally affect the market. But the Zestimate goes beyond estimating the price of a home, it creates a price by influencing the beliefs of buyers and sellers about what a home is worth. In this way, Zestimate has been acting as a market maker even before Zillow got into the business of buying homes based on its estimates.

While the innovative potential of these technologies is enormous, it is always important to consider possible systemic and second-order effects, as well as potential unintended consequences, when introducing machine learning dynamics to the market. On the academic side of this issue, leading AI researcher Michael I. Jordan paints a picture of the synergy between recommendation engines and marketplaces. In particular, he points out the harm done by naively applying recommendation engines to scarce resources. For example, recommending an item of limited supply to many shoppers artificially inflates the demand for an item that may be an alternative or even superior to some buyers. Likewise, a navigation app that recommends low-bandwidth shortcuts for many drivers can create serious congestion.

The essence of AI algorithms is to compress rich information into a concise form, but the market is instead a generator of rich information because it exploits the heterogeneity of a broad buyer base. If AI algorithms are allowed to dominate the system, it will become obsolete, as strong recommendation engines reduce product variety, and loss of consumer choice reduces the market's ability to elicit preference information from consumers. We can create better systems through rigorous testing and design validation during the engineering process.

ARM classes and instances

As we continue to work on ARM models, we need to be clear about our goals. Our goal is to balance the compression power of machine learning with the discovery power of markets to facilitate new forms of high-dimensional energy markets (among other use cases). As we mentioned in our fungibility paper, this is best understood formally in the context of online learning. Our work on economic games as estimators and constant function market makers (CFMMs) as oracles shows that certain types of intelligent algorithmic pricing models can be interpreted as signal processing operations that learn how markets price goods and services ( Goods in these cases) - "Contextual Goods".

The way forward for ARM R&D requires us to think of it as a "class" where each "instance" fits into its specific domain. In machine learning, this includes model selection, feature engineering, meta-parameter optimization, ensembles, and other customizations applied by data scientists to fit a particular model to a particular problem domain. For AMMs, this includes choosing a specific design pattern, such as Uniswap, Balancer, Curve, etc. These different CFMMs are characterized by different underlying mathematical invariants. But even after choosing a class of AMMs, there are still some unique instances with their own assets and other meta-parameters, such as fees and weights.

It must be admitted that this class/instance relationship is a cutting-edge extension of ARM research. Because of its similarity to ML, we understand that the difficulty of developing and maintaining an ARM can vary widely, depending on the information the model is synthesizing. BlockScience is collaborating with Hedera Hashgraph and the HBAR Foundation's Sustainability Initiative to design an ARM instance specifically for the energy credit market, as energy credit assets provide a good use case for projects with multiple attributes and value is contextualized great impact.

An example of ML data clustering by attributes. In ARM, clusters like this could represent renewable energy credit attributes and be used to facilitate the "match" of supply and demand environments for semi-fungible assets. (Source: https://en.wikipedia.org/wiki/Cluster_analysis)

secondary title

Why energy and carbon markets are great use cases for ARM

Currently, indistinguishable energy credits are bought and sold in bulk with little transparency and auditability. The renewable energy market is also highly artificial, and carbon offset/removal credits and renewable energy credits traded by brokers can have very different contexts. Although these assets are traded as commodities (fungibles), in reality their value in the market can vary widely depending on where and how the energy is produced, the means of production, the quantity and quality of the equipment used, and their attributes and how these properties are evaluated in different contexts.

With ARMs, these assets can be more granularly segregated, as the market or demand side can dictate different prices for these semi-homogeneous assets. Buyers of energy assets require more information and visibility on the supply side, often based on legal requirements in different jurisdictions. Those involved in these markets want to understand which properties are highly sought after and which are more common in order to exercise buying power while meeting requirements.

In ARM, semi-fungible goods—"bundles" of various attributes—have different weights depending on market conditions and production circumstances. This would provide better information for the market to match between supply and demand parameters - in other words, a more efficient market. Additionally, this functionality can provide value by discovering energy and carbon credits on a public ledger and linking them to their auditable origin through verification.

Ensuring that a CORC or REC is unique and represents a carbon offset/removal on-chain or an energy asset created off-chain is a very important aspect of this market innovation. A core part of the Hedera ecosystem and the HBAR Foundation's sustainability initiatives is Guardian, a fully auditable solution that verifies the attributes of energy assets. Guardian3 provides proof of quality for off-chain data, including decentralized identity, policy-driven actions, and fair ordering of transactions. These properties provide a means of eliminating well-recognized data quality issues such as double counting of assets, which confound supply and demand, and will be at the heart of research and development in the automation of CORC and REC using ARM.

Potential Benefits and Global Impact of ARMing Energy Market

The global energy-as-a-service market size is expected to exceed USD 106.6 billion by 2026, and with the growing importance of CORC and REC, these markets are ripe for innovation. The ARM mechanism could have a huge impact on the energy credit market and open up a new world of possibilities by connecting the technology of automated market makers with real-world variable assets represented on-chain.

ARMs can provide automation, allow massive scaling and expansion of energy markets and trading, accelerate market development and provide the potential for futures markets. Costs can also be reduced for individuals, smaller energy producers or cooperatives, as they can more easily aggregate decentralized generation networks to provide collateral or tokenized assets in exchange for financing. With more efficient and expanding markets and greater visibility into energy production and attributes, there will be greater incentives to improve and invest in infrastructure that meets higher or more valuable standards or standards required by the market. Greater visibility into localization can also keep the value of local production within these boundaries rather than being extracted to larger markets.

However, as with any technology or innovation, there are many unexplored knock-on effects – any new frontier brings with it the potential for unintended consequences and the systemic impact of new market mechanisms. As mentioned earlier, our goal is to strike a balance between the compression power of ML and the discovery power of the market. As an engineer, there are always design tradeoffs. That's why modeling and simulation work is extremely important, and why we use cadCAD to optimize complex system designs.

Visualizing High Dimensional Spaces Using Machine Learning (Source: https://www.youtube.com/watch?v=wvsE8jm1GzE )

secondary title

Modeling ARM with cadCAD

cadCAD (short for Complex Adaptive Dynamics Computer Aided Design) is an open source modeling framework for studying, validating, and designing complex systems using computers. This helps individuals or organizations make informed, well-tested decisions about how best to modify or interact with complex systems to achieve their intended goals.

2) Enforcing accounting rules, for example, to ensure that adversaries cannot simply bankrupt ARM by flooding it with unsellable energy credits.

refer to

At the same time, the Hedera ecosystem is developing the core application logic needed to deploy ARM, execute business logic and accounting on the Hedera public ledger, such as their recently open-sourced Guardian for validating on-chain claims against real-world actions.

While we will develop the technical requirements to support the deployment of the Energy Credit ARM, the results obtained will generalize to a wide range of partial alternative, two-sided market applications on its infrastructure. The Hedera ecosystem not only focuses on deploying fully custom ARM instances, but also creates ARM application design patterns for its infrastructure. This will allow applications to focus on the data science aspects of their ARM design and trust their market design to the deployment and implementation of the Hedera network.

//www.youtube.com/watch?v=842acSWmBC4&t=1093s

In the future, topics such as secondary effects, system-level incentives, and other underlying parameters of the mechanism could be investigated to gain insight into how the ARM "learns" and operates.

2. Oladunni,secondary title

Shorish,refer to

3. https://github.com/hashgraph/guardian