Risk Pod Creation Proposal

Summary

This proposal asks to establish the Risk Pod within Gro DAO. The proposal will be open for comments over the next 3 days. If no substantial changes are required, it will then be open for voting for another 5 days.

Facilitator

@pavel

Mandate

In its activities Risk Pod proceeds from the premise that a web3 protocol is an evolving shared narrative living simultaneously in two domains open for exploration: onchain transactions and community interactions. They inform and impact each other in both directions in a myriad of ways creating a bipartite heterogenous living organism. To be able to evaluate risks stemming from Gro DAO’s involvement in web3 means to be able to model the evolution of both Gro protocol and its web3 counterparties in a coherent, unified way.

With the premise above in mind, the Risk Pod Mandate includes but is not limited to the following tasks:

  • Risk Parameter Proposals
  • Yield Strategy Whitelisting and Delisting Proposals
  • Monitoring of Portfolio Exposure
  • Community Involvement
  • Monitoring DeFi Systemic Risk
  • Research
  • Risk Tooling

DeFi Systemic Risks

Gro DAO is highly integrated in DeFi which continues to expand in total value locked and its complexity. In the long run Risk Pod team needs to evaluate how any potential parameter change or new yield strategy whitelisting at Gro DAO might affect systemic imbalances. Furthermore, it needs to monitor DeFi to properly address liquidity and other risk related concerns.

Research

Besides ongoing risk related work and tasks, the team also aims to perform regular research on potential new protocol design topics and how it might benefit Gro DAO’s risk profile. This also includes monitoring of DeFi addressed above and considering potential integrations with other protocols.

Risk Monitoring & Tooling

Risk monitoring is ensured by regular usage of our internally developed risk models. All risk estimates and methods used are being reported openly.

Plan is in the long run to have all our risk models completely automated and run regularly with an open web access so that the community can get informed about Gro’s risk profile at any time.

This also allows easier access to new Risk contributors and establishment of additional Risk Pods.

Note that Risk Pod is focused on market risks of the protocol, whereas other types of risks should be addressed by other pods with corresponding mandates.

Workforce

For the foundational period in addition to the Facilitator the team of Blockanalitica is planned to be involved as a primary contractor. Blockanalitica aka MakerDAO Risk CU will be able to leverage their extensive experience as well as battle-tested codebase from their work on MakerDAO risk modelling tooling and methodology as well as Aave analytics dashboard.

Budget

The next 3 months will be foundational in the sense that Risk Pod will deliver basic dashboard and qualitative analysis tooling for Gro DAO risk modelling as well as methodology of the results translation into Gro DAO Governance proposals in the style of Computer-Aided Governance.

In particular, it will include risk modelling tooling as a customised version of Blockanalitica software.

It will also include a model for basic analysis of the DeFi community sentiment towards key crypto assets used in the Gro ecosystem leveraging the results of @pavel’s previous DeFi sentiment risk modelling (see the UST sentiment => crash case in particular) and will deliver an improved BERT_large language model, 355m parameters (vs BERT_base, 135m parameters, used in the UST/LUNA demo case).

I propose a budget of 140,000 USDC for the next 3 months, in particular:

  • 70k for the Blockanalitica customised tooling;
  • 30k for the sentiment analysis tooling;
  • 10k for the methodology;
  • 5*3 = 15k for the Facilitator 3 month compensation;
  • 15k for the Contingency Buffer.

that will be transferred to the Risk Pod operational wallet if approved.

Reporting

Risk Pod will update the DAO on its progress regularly through the community channels such as Community Forum and Discord including a 3 month report on the Community Forum.

Domain Evolution

We believe the best way to scale Risk at Gro DAO is to have emergent structures from an initial one proposed here. Risk Pod scaling should be ideally done in a way to have “risk field specialized units” separated from the initial one with its own facilitator.

For instance, a team member within the Risk Pod may want to specialise on Labs products. He creates his own Labs Risk Pod with his own team and budget. He would be separated from the initial Risk Pod, but would still collaborate with it.

In our opinion, such evolution of teams within one broad domain such as Risk is preferred because it doesn’t lead to work overlapping and cost inefficiency, which would be the case if we were to have multiple Risk Pods performing the same type of work initially.

We do however support development of separate additional Risk Pods and are willing to collaborate with them. We only believe development of such units may be most efficient when developed from initial teams where common language already exists and coordination is easier.

If I understood correctly, a portion of the budget is dedicated to research work of a custom DeFi sentiment analysis software.

have you done studies to prove that twitter sentiment is a sufficient predictor of DeFi protocol risks? How do you expect the work developed here to be used by the protocol?

sure, already in 2010 it was shown

that the accuracy of a model predicting daily closing values of Dow Jones Industrial Average is significantly improved by the inclusion of specific public mood dimensions as measured from Twitter feed.

Most recently it was confirmed

that similar impact is present in the case of Twitter sentiment and crypto market.

Speaking more generally, Nobel-winning research by Robert Shiller shows that irrational factors like sentiment and narratives play an important role in asset pricing. In fact, the Nobel Prize in Economics ’13 was shared by Shiller and Eugene Fama, the creator of the Efficient-market hypothesis. This underscores that EHM- and sentiment/narrative-driven asset pricing are two complimentary mechanisms and market risk modelling should be handled accordingly.

The most obvious example of irrational exuberance on financial markets are bubbles like the one happened around UST — all information about its model was open, yet the bubble emerged and then collapsed. Such price movements also happen gradually over significant periods of time. Price volatility not explainable by objective economic factors happens in case of mature assets like ETH as well, see e.g. crypto winters. It’s also an example of risk beyond the scope of risk models based on the Efficient-market hypothesis.

Nobel Prize in economics ’22 was awarded for showing mechanisms of bank vulnerability to rumours becoming a self-fulfilling prophecy and leading to a bank collapse (bank runs). This once again underscored the importance of financial markets sentiment analysis.

Estimates of market irrationality impact

It was shown that only about 7% of the variance of annual stock market returns can be justified in terms of efficient market factors for the aggregate U.S. stock market 1871–1987. For individual-firm stock price variations it was estimated that the standard deviation of the ‘‘atypical discount’’ was about 25%. It was concluded that the inefficient component of stock price variation that generates predictable movements in future returns ‘‘still has an economically significant impact on firm-level stock prices’.

Importantly, given high correlation between DeFi assets and due to the impact of sentiment/narrative on the DeFi market via social media/social networks like Twitter and Discord, it’s very likely that the impact of inefficient market factors on DeFi asset price movements is far higher than that for individual TradFi asset prices and closer to 50%, maybe more. See crypto bubbles/winters.

TradFi is already in

Moreover, The Federal Reserve Bank of New York has been gauging public sentiment since 2013 (10) and integrating this data into its models.

Baker and Wurgler reviewed the investor sentiment literature and showed that a “sentiment index” is highly correlated with aggregate stock returns.

See also Handbook of Economic Expectations and The overview of literature on expectations data in asset pricing (14) in particular.

See also R. Shiller’s book on Narrative economics. (15)

See also Nobel Prizes:

’17 “for exploring the consequences of limited rationality, social preferences, and lack of self-control … how these human traits systematically affect individual decisions as well as market outcomes.”

’02 “for having integrated insights from psychological research into economic science, especially concerning human judgment and decision-making under uncertainty”

At the first stage Sentiment dynamics will be wrapped into the dashboard as a tool for Computer-Aided Governance: a qualitative factor to be considered when proposing protocol parameters for the Gro DAO vote.

Then it will gradually be integrated as an input into the quantitive Market Risk Model. One approach: there are two components: EMH and non-EMH. EMH is modelled as Geometric Brownian motion. It’s augmented by a non-EMH component modelled with a Sentiment Model.

Once enough time-series data is accrued a regression model could be learnt with sentiment + vault activity as input and Risk Premium prediction as output.

The eventual goal could be for protocol parameters becoming fully autonomously derived from onchain and crypto community/offchain data.

I would counter that our market is far to immature for AI sentiment. The models indicated were based on over a decade of consistent data on a mature, massive market. Our market is not yet any of those things. We, as an industry do not have an agreed risk model, do not have an agreed method for risk communication. We need to build a foundation (that TradFi has been optimizing for a decade) before we try these complex models. Even applying the Bockanalytica data may not add value. The transaction numbers for Maker are much higher than Gro.

1 Like

Block Analitica doesn’t provide data. They’ve built dashboards for Maker and Aave for risk monitoring in DeFi.

What models do you mean? Sentiment plays significant role on any financial markets because human nature is the same. It’s obvious both for the economic theory, behavioural economics in particular, and from the common sense PoV: UST crash case, crypto winters etc. are examples.

As for gauging sentiment of the DeFi community, Large Language Models, which is the state-of-the-art tool here work with language and text data, not markets.

Imo, foundation building for the industry is happening just like that: experiments, trial and error, incremental steps. Grand planning doesn’t work in practice usually. Maker’s battle-tested platform for risk management in DeFi is an excellent place to start for Gro standing on the shoulders of giants.