Dynamic modeling for optimal cryptoeconomic policies 

| tokenomics | monetary theory | dynamic optimization |

(see github repo)

This research project applies dynamic stochastic general equilibrium methods to model and optimize crypto policies for staking and burning. My goal is to provide insights for designing a cryptoeconomic system that self-regulates without requiring governance power.

Currently in the phase of model formalization.

Deep Learning for Dynamic NFT Valuation

| NFTs | deep learning | DeFi | zkML |

(see arxiv paper and github repo)

Current NFT-Fi applications face bottlenecks because it is extremely difficult to price NFT collaterals fairly and accurately. In this paper, I showed that this issue can be solved via deep learning and ZKPs.

I trained a convolutional neural network based on historical trades, market trends, and traits/rarity data of BAYC NFTs (sourced from Dune and OpenSea). Then I proposed an application framework using zkML techniques to provide a fully trustless price prediction oracle for NFTs in any onchain collection.

Large-scale time series modeling for Ethereum transaction fees with AWS and PySpark ML

| on-chain analysis | cloud computing | machine learning |

(see github repo and video presentation)

I built a time series machine learning pipeline for Ethereum transactions fees. I was able to analyze a large set of data (~500 million raw transactions) by leveraging the processing power of AWS EMR clusters and an efficient Spark framework. 

The end-to-end pipeline consists of custom wrappers built on Spark ML's transformer class. I wrote custom functions for train-test-splitting and hyperparameter tuning and got significant improvements in forecast accuracy over the baseline model.

Benchmarking multi-agent reinforcement learning models with evolutionary games on complex networks

| distributed systems | network analysis | game theory |

(see published paper and conference presentation)

How did we selfish humans learn to cooperate and form stable social structures? What happens if some of us don't cooperate?

I led a student research group to study the phenomenon of network cascading failures induced by agents' failure to cooperate in the Prisoner's dilemma game. We studied how network topology and evolutionary parameters affect cooperative behaviors.

My main contributions include designing and developing multi-agent systems in Python using networkx; conduct controlled experiments using computer simulations; analyzed and visualized data with pandas.

A quantitative analysis of Maker (MKR)

| quantitative finance | DeFi |

(see research report and thread)

I took a quantitative research approach adopted from Liu and Tsyvinski (2021) to study MKR returns on different time scales. 

I analyzed the predictive power of various factors including other crypto, momentum, macro indicators, Internet search trends, and network fundamentals. Then I discussed an effective trading strategy using a combination of technical and fundamental signals.

numerics package and gradient descent

| numerical analysis | gradient descent | Python |

(see github repo and research paper with math proofs)

I wrote from scratch a Python package containing 18 numerical algorithms for root finding, matrix decomposition, interpolation, and optimization.

Then I studied the gradient descent algorithm from a numerical analysis perspective. Highlight: an independent proof for the existence of an optimal step size under the Wolfe conditions.

I illustrated my gradient descent algorithm with constrained utility maximization problems with the Lagrange multiplier method.

Game theoretic models of the zero-commission     brokerage war

| game theory | finance |

(see short paper)

I proposed three game-theoretic models for the strategic interactions in the zero-commissions war among major U.S. stock brokerages:

• A repeated Prisoner's Dilemma model

• A collective-action model

• An asymmetric-information signaling model