# Treasury Futarchy *Decision markets for treasury allocation.* *Conditional Funding Markets* --- ## Context ### customers - foundations: stewards of governance and treasury for DAOs - project leaders: seeking funding from DAOs - tokenholders - community and speculators --- ### example: Optimism's Citizen House - [Dual governance](https://community.optimism.io/docs/governance/): tokenholders + Citizen House - Citizen House acts as a jury --- ### retro funding rounds [ref](https://gov.optimism.io/t/upcoming-retro-rounds-and-their-design/7861)  --- ### metrics [ref](https://models.opensource.observer/#!/model/model.opensource_observer.rf4_impact_metrics_by_project#description)  --- ### opportunities for improvement - **eliminate capture by jury** <- elicit information, not preferences - **enable funding under-funded and risky projects** <- elicit predictions - **counterfactually compare allocation decisions** <- compare conditional predictions --- ## CFM [Conditional Funding Markets](https://community.ggresear.ch/t/conditional-funding-markets/27) --- ### setting metrics $m$ with weights $w(m)$, defined by the foundation  --- ### setting (cont'd) - budget $b$, defined by the foundation - investment ask $i(p)$, defined by each project  --- ### objective **we want to allocate funds to projects so as to maximize metrics-based return on investment**  --- ### overview - elicit $\sum_{m} w(m)\cdot m(p)$ - at some future date - conditional on *Funding* or *No Funding* - allocate fund to: - maximize ROI $\frac{\sum_{m,p} w(m)\cdot m_{\text{actual}}(p)}{\sum_{p} i_{\text{actual}}(p)}$ - respect a budget $b$  --- ### components per slot: - steps: - 1 auction - mutiple conditional prediction markets - 1 allocation algorithm - budget: - split into allocation vs liquidity subsidies - proportions based on market conditions (centralized at first) --- ### step 1: auction project leaders bid for inclusion, first-price - bids: - amount - investment ask $i(p)$ - initial predictions for the Funding and No Funding case --- - selection rule: - order by amount / investment ask - compute complementary liquidity subsidies required - limit by liquidity budget *<- this should include more winners than the allocation budget* - winners gain access to next step - revenue allocation: allocates to AMM liquidity using input predictions --- ### step 2: conditional prediction markets 2 prediction markets per project: - scalar markets predicting the weighted sum of metrics - conditional on Funding ($i(p)$) or No Funding - CPMM, liquidity bootstrapped with auction bid + liquidity subsidiy - initial predictions used as initial price - subsidy withdrawn after 1 week, outcome value aggregated over several days - markets run until either cancellation or resolution --- ### step 3: allocation algorithm - allocate funding to projects: - naive algorithm: - ordered by individual ROI $\frac{\sum_{m} w(m)\cdot (m_{\text{Funded}}(p) - m_{\text{Not Funded}}(p))}{i(p)}$ - limit by allocation budget - allocate at random some of the time, to prevent manipulation - resolves Funding / No Funding conditionals, cancels markets --- ## Remarks - auction: spam-prevention & shifting PM risk towards projects - some randomization appears acceptable as far as treasury allocation is concerned - tradeoff to be studied further - (Othman, Sandholm) counterfactual requires randomization - other, non-counterfactual-comparison-based mechanisms to explore - do we still achieve objectives?