New tools for backtesting and open-source quant stacks

Backtesting is the replay tape of trading: you run a strategy on old market data to see every trade, gain and drawdown before risking cash. This week, a cluster of new open-source projects and one new application programming interface pushed that workflow closer to plug-and-play. (signal-sigma.com) Signal Sigma said on April 22 that it had brought a `/backtest_result` endpoint online, letting users pull “every detail” from Strategy, Measure-Trace and Signal-Trace backtests programmatically. Its API site says endpoints require bearer-key authentication and publishes both interactive docs and an OpenAPI schema for code generation. (signal-sigma.com) (signal-sigma-api.com) (signal-sigma.com) The company said the endpoint returns the same artifacts researchers usually inspect by hand: snapshots, trades and equity curves. In practice, that turns a backtest from a static chart into a machine-readable record that can be routed into dashboards, screening jobs or live research systems. (signal-sigma.com) The open-source side of the stack is moving at the same time. `ai-hedge-fund`, a GitHub project described as “a proof of concept for an AI-powered hedge fund,” had 56,100 stars and 9,700 forks when GitHub was crawled this week, while TradingAgents was presented as a multi-agent financial trading framework for research use. (github.com 1) (github.com 2) TradingAgents’ public site says the framework splits the work of a trading desk into specialized roles, including fundamental, sentiment and technical analysts, plus researchers, traders and risk managers. The project warns that results vary with model choice, temperature, trading period and data quality, and says it is not investment advice. (tradingagents-ai.github.io) (tauricresearch.github.io) Kronos tackles a different layer: the market model itself. Its paper and code describe a foundation model for financial candlesticks, trained on more than 12 billion K-line records from 45 global exchanges, with an autoregressive transformer that treats price sequences more like language tokens than spreadsheet rows. (arxiv.org) (openreview.net) (github.com) A separate GitHub repository, `kronos-hub`, surfaced this week as an integration layer that connects Kronos forecasts with TradingAgents research and AI Hedge Fund execution and backtesting. The repository describes itself as an “integration-first API hub” and says a hybrid demo can fan signals out into those other tools when credentials are available. (github.com) The common thread is formatting. When model outputs, agent decisions and backtest records all arrive in standard fields over an application programming interface, engineers spend less time writing one-off adapters and more time comparing strategies on the same rails. (signal-sigma.com) (signal-sigma-api.com) (github.com) That does not settle whether any of these systems make money. The open-source projects frame themselves as research tools, and TradingAgents and ai-hedge-fund both present their work as simulation or research rather than live investment products. (github.com 1) (github.com 2) (tauricresearch.github.io) What changed this week was not a new trading law or a new exchange feed. It was the growing availability of off-the-shelf parts — model, agent framework and backtest API — that let a small team assemble an automated research stack much faster than before. (signal-sigma.com) (github.com 1) (github.com 2) (github.com 3)