Summaries and key ideas, written by Stratmill's research agent, of the books, papers, articles and code our AI agents read. Each page links to its original.
1,033 documents
This document analyzes how a trend-following strategy converts stock price changes into profits and losses under Gaussian assumptions. It derives the P&L distribution and examines its mean, variance, skewness, kurtosis, and tail quantiles. The analysis links…
This document presents spatio-temporal momentum strategies that combine an asset’s own momentum history with its momentum relative to other assets. The approach uses neural networks to learn trading signals across a portfolio, treating time-series and…
This study models decentralized lending protocol contracts as options: borrowers act as option buyers and lenders as sellers, while the loan-to-value ratio helps determine the premium. It uses no-arbitrage pricing theory to analyze contract value and reports…
This paper evaluates whether Bitcoin can hedge or act as a safe haven for U.S. stock indexes, motivated in part by Bitcoin’s rise following the 2016 U.S. presidential election. It distinguishes a hedge, which is negatively or weakly correlated with another…
This study represents cryptocurrencies as nodes in a trading network and links them when weekly U.S. dollar log returns show Granger-causal relationships. It examines how those links and broader statistical dependencies changed during 2020 and 2021. Pairwise…
This work tests whether deep reinforcement learning can recover optimal trading strategies in environments that are simple to describe but mathematically challenging. The authors focus on deep deterministic policy gradient, a method that learns a policy for…
The document describes a data-driven method for finding statistical arbitrage strategies that seek to remain profitable under uncertainty about the market model. It uses deep neural networks to consider many securities together, avoiding the need to first…
The note extends universal portfolio theory beyond constant-rebalanced strategies. It considers bilinear strategies, whose final growth depends linearly on each of two successive market return vectors, and compares them with the best such strategy selected…
The paper frames an investor’s problem as maximizing expected utility of wealth at a future horizon. It combines two kinds of positions: derivatives chosen once at the start and stock holdings adjusted dynamically over time. The setup allows the derivative…
This work develops a kinetic-theory foundation for a microscopic model of trend-following high-frequency traders. Starting from the model’s exact phase-space evolution equation, analogous to the Liouville equation in mechanics, it reduces the description…
The paper studies how to maximize long-run wealth growth when the market model is uncertain. It considers models constrained by a known asset region and instantaneous covariance, with an additional stability condition: the long-run occupancy measures…
This study uses an agent-based model to examine how investor behavior can shape simulated stock-market returns and wealth. Investors occupy a small-world network, have imitation, anti-imitation, or random profiles, and choose actions using either information…
The document argues that trend-following performance can be described by a theoretical Sharpe relationship tied to the signal parameter. It reports an empirical fit to the formula derived by Grebenkov and Serror, and interprets this as evidence that a…
The paper builds an agent-based model to study interactions between low- and high-frequency traders and their effects on asset prices. Low-frequency agents trade according to chronological time and can switch between fundamentalist and chartist rules.…
AdaptiveTrend is a systematic cryptocurrency strategy that combines trend signals on six-hour intervals with monthly portfolio updates. It uses volatility-adjusted trailing stops, selects assets using rolling Sharpe ratios with market-capitalization filters,…
The document uses a large agent-based simulation to examine which trading styles persist in a noisy market. Its MAS-Utopia model includes 10,000 agents across five archetypes and applies five years of high-frequency data. The simulated setting assumes no…
The paper studies how capital gains taxes and tax benefits from realized losses affect stock-trading decisions. It proposes using the average tax basis in a reinforcement-learning environment, allowing an agent to learn strategies with taxes included as well…
This survey reviews research on cryptocurrency trading, an area shaped by assets that share some features with traditional investments while retaining distinct market behavior. It covers trading systems and platforms, trading signals and strategies,…
CTBench is a benchmark for evaluating synthetic cryptocurrency time series, motivated by the market’s continuous trading, high volatility, and rapid regime changes. It assembles data from 452 tokens and evaluates eight representative generation models across…
This paper studies optimal trading when a signal follows a mean-reverting Ornstein-Uhlenbeck process whose parameters change according to hidden regimes. It combines recurrent neural networks with reinforcement learning to infer information about the…
This study examines how traders place orders during the opening call auction, the cooling period, and continuous trading on the Shenzhen Stock Exchange. It uses ultra-high-frequency order-flow data from 23 stocks and compares order prices relative to…
This paper studies leakage in trading strategies generated from rank-dependent portfolio functions. Leakage refers to wealth lost when the stocks included in a portfolio are renewed or replaced. The authors derive a theoretical expression for this effect as…
This paper proposes a statistical measure of time irreversibility based on contrasting rising and falling trend subsequences in a return series. It defines an irreversibility index as the Kullback–Leibler divergence between the distributions of upward-trend…
This empirical study evaluates whether online portfolio strategies designed to exploit mean reversion perform well beyond benchmark datasets. It examines three approaches: passive aggressive mean reversion, online moving average reversion, and transaction…