Reprinted from Jianlin Su’s article “An Introduction to the Median“. The annotations in this post were added by me.

I recently revisited the concept of the median, so I decided to write down the key points while they are still fresh.

When removing or clipping outliers, we often need a “baseline.” For example, for a batch of nonnegative data, we may regard any value greater than 50 times the baseline as an outlier. So how should that baseline be chosen? A common choice is the mean, but the mean is easily pulled away by outliers. As a result, using it as the baseline may bias us toward those outliers and cause some problematic results to be missed. In such cases, the median can be a better baseline.

As traditional dual-low strategies gradually lose their effectiveness, what else can be done in convertible bond investing? This article introduces a complete quantitative framework for convertible bonds: using equity multi-factor Alpha to drive security selection, using the CCB pricing model as the strategic hub linking returns and risk control, and finally using stock index futures to hedge Beta.

In the discussion in the previous article, “AMM Quant Deep Dive 01”, we used Ito’s lemma to show that the core cost borne by an LP is LVR (Loss-Versus-Rebalancing), which takes the form \(\frac{\sigma^2}{8} V dt\) under the constant-product formula. That result revealed the essence of the “volatility tax,” but to hedge risk more precisely, we still need a tighter pricing framework.

Drawing on recent research by Hou, Singh, Echenim, and others, this article reviews the static structural prototype of LP positions and discusses their risk and value characterization from two angles: absolute valuation under boundary stopping times, and marginal cost under a continuous-installment perspective.

This is the full story of a Solana on-chain strategy, from the investment thesis to Kalman-filter-based hedge modeling, convex optimization for the optimal portfolio, and aggressive leverage through looping loans, ending with half the profit given back after the bull market tide receded.

In derivatives pricing and large-scale risk backtesting, where every millisecond matters, traditional normal-distribution integration often becomes both ugly code and a performance bottleneck. This article introduces a fully vectorized numerical algorithm that lets you exploit the parallelism of modern CPUs, GPUs, and even TPUs to break through that bottleneck.

Many people treat providing liquidity (LP) on on-chain exchanges (DEXs) such as Uniswap as a passive form of “yield management” or “mining.” That is an extremely dangerous misunderstanding. From the perspective of market microstructure, an AMM is fundamentally an adverse-selection game arena defined by an algorithm.

This article is the first entry in the [AMM Quant Deep Dive] series. We will abandon the traditional retail-investor perspective, start from macro market equilibrium, strip away the disguise of “impermanent loss (IL)” through rigorous mathematical derivation, and introduce the true core risk metric for professional AMM market making: Loss-Versus-Rebalancing (LVR).

In the earlier article Three Measures for Sequence Correlation, I introduced Chatterjee’s correlation coefficient, which can detect nonlinear and non-monotonic associations between variables. Its form is simple, computation is fast, and it is especially suitable for screening input features for nonlinear machine learning models. New research, however, shows that its detection efficiency is inadequate for some non-functional dependence structures. This article first reviews and expands on the concrete mechanics and key properties of Chatterjee’s \(\xi\). It then introduces an alternative: the equally consistent, distribution-free independence test statistic \(\tau^*\).

In the crypto world, the decentralized exchange (DEX) is a beautiful ideal: it allows anyone to custody and trade their own assets securely, without permission and without trusting a third party. But on the road to realizing that ideal, the space has long faced a painful trade-off.

When things reach an impasse, change becomes possible; through change comes flow; through flow comes endurance.

The techniques of markets change without cease. High-frequency market making chases light at the microsecond scale; on-chain protocols build pools of liquidity with mathematics; spot-futures arbitrage searches across different time horizons for the pull of equilibrium. There are countless methods.

And yet the Great Way is simple, and all methods return to one origin. There is a path that does not cling to technique but wanders through principle, turning complexity back into simplicity: the practice of the “generalized liquidity provider” (LP). Those who follow it eventually see that market making and arbitrage are merely two sides of the same coin, each rooted in the other.

These notes are a record of that practice. What they seek is not a bag of tricks, but a way that runs through the whole system: a unified underlying framework for every form of liquidity provision. Only then can one build durable understanding in a market that never stands still, and ultimately move in harmony with it.

This report systematically reviews the key links in the low-latency trading stack for China’s A-share market in 2025 and beyond, from exchanges to brokers to the client side. Achieving extreme low latency requires coordinated optimization across trading units, ultra-fast broker counters, low-latency brokerage and market data services, networks, algorithms, and hardware. New regulatory rules also raise the bar for both technical implementation and compliance. By combining the latest technology and policy trends, the report evaluates the latency characteristics of different approaches and provides a reference for market participants.