Four new sections: market regimes, claim checker, event replays, famous strategies
2026-11-11
Market regimes shows each market's trend, volatility and strategy positioning every trading day. The claim checker tests 34 popular trading sayings on daily data and gives each a verdict. Event replays show how the strategies behaved in 35 major market events since 2010, plus recent sell-offs detected automatically. Famous strategies re-runs 22 well-known public portfolios and rule sets on our ten markets and says what had to be adapted.
Optional coffee support for continued research
2026-11-08
An optional Ko-fi link now appears in the footer, at the end of research articles and beside prepared or verified browser downloads. Contributions support data updates, backtest verification and new tools. All existing free downloads remain available. The support provider opens only when you choose the link; no payment widget loads automatically.
Compare portfolio additions and removals from pinned evidence
2026-11-07
A new comparison worksheet keeps before and after on identical shared dates, separates strategy and allocation fees, and shows gross/net returns, recovery, downside sample counts and signed exposure overlap. Frozen share links retain the original evidence; optional named saves stay in this browser. Independent sleeves do not net trades across strategies. Financing, borrowing, funding, taxes and market impact remain excluded.
Cost comparisons across 576 supported pairs
2026-11-07
All 576 supported strategy-market pairs now have separate 0, 5, 10 and 25 bp cost scenarios across seven windows. Strategies and matched benchmarks each pay charges on their own drift-adjusted one-way risky turnover after gross returns. Rankings compare identical asset universes, dates, input returns and window anchors; a basket labelled SPY is kept separate from single-market SPY. Fee totals use the disclosed initial-capital or start-close-capital units. Metrics use all original full-precision daily records, while curves are bounded display samples and metrics load by the selected universe. This independent fee-sensitivity view leaves existing before-cost Arena rankings intact. It models synthetic exposure charges, not actual broker fills or exact traded notional; financing, borrow, funding, taxes and market impact remain excluded.
Common-date AI season scheduled after public freeze
2026-11-07
OpenAI and Anthropic each retain six unchanged rules. Their prompt, original replies and receipt metadata were verified publicly on 7 November 2026 at 00:20 UTC, before scheduling. Season 1 runs from 9 November 2026 through 6 February 2027. It is scheduled, with no performance yet; rankings require at least 20 observations in every required market and complete matching windows. Late entrants belong to a later season.
Common-date AI seasons: two frozen provider submissions
2026-11-06
Two real CLI submissions, from OpenAI and Anthropic, now retain all six frozen rules each under one public prompt, with exact prompt/reply downloads, requested and reported model fields, freeze timestamps and operator-recorded receipts. The common-date season remains a draft: no start has been scheduled and no performance or rankings are published. Anthropic reports claude-sonnet-5 and discloses auxiliary claude-haiku-4-5-20251001 usage; the auxiliary purpose and underlying model revision are not established, and inference compute is not normalized across providers. The original single-assistant experiment remains separate with its original figures and disclosed execution-timing and rounded-input limitations. These CLI receipts are not provider-signed attestations.
Full-cohort evidence and causal calendar corrections
2026-11-06
Evidence now covers all 72 strategies and 576 supported pairs, including signed per-asset basket holdings. Daily records, configurations and archived source can be downloaded on demand; the standalone tool can fetch inputs and regenerate rules when the original input fingerprint matches. Every current run has original-input recomputation proof across 2,394,468 observations, including 11 newly corrected runs. Ten monthly basket rules now decide on the first observed new-month session using the preceding observation, with the engine lag retained. The turn-of-month rule uses a disclosed weekday approximation, not an exchange-holiday calendar. These corrections change historical results; old immutable runs and correction links remain available. Costs remain zero in this baseline; synthetic targets and rebalancing are not broker fills.
Auditable strategy evidence and direct engine positions
2026-11-06
Six representative strategies now have evidence bundles on SPY, gold and bitcoin: full-precision daily accounting, simulated position changes, actual selected parameters, source hashes and a standalone arithmetic replay script. The evidence index preserves prior corrected snapshots. Signals now come from engine targets rather than ratios of rounded returns, with target and effective exposure, per-market observation dates and independent data status. The current UTC date is excluded conservatively. Walk-forward boundary returns are recomputed from the previous published target, and skipped folds retain default targets; the evidence index quantifies these accounting changes. This is an 18-pair pilot, not full-site verification or broker fills.
New: AI Strategy Arena
2026-11-04
Six trading rules written by an AI assistant, with every parameter fixed, are now tracked every trading day on nine markets against buy-and-hold and our 72 classic strategies. Because a language model may already know past markets, the ranking uses only results after 4 November 2026, the day the rules were frozen; the historical backtest is shown for reference only.
New: open-source trading frameworks directory
2026-11-04
A guide to 18 widely used open-source projects for backtesting, live trading, crypto bots, machine learning and market data (Backtrader, vectorbt, LEAN, NautilusTrader, vn.py, Freqtrade, Qlib, FinRL, TradingAgents and more): what each is for, strengths, limitations, alternatives and seven head-to-head comparisons. GitHub stars, last commit, latest release and maintenance status are refreshed every week.
Subscribe to the weekly report
2026-11-04
Every static page now has a subscribe box for the weekly Arena report (ranking changes, new signals and new research), sent by email through Buttondown. We also started counting anonymous usage events such as searches and comparisons with Umami, which uses no cookies.
A site map for AI assistants (llms.txt)
2026-11-04
The site now publishes llms.txt, a plain-text guide for AI assistants and answer engines: what Quant Arena tests and how, the main pages, every live research article with its data date, the glossary and all strategy pages. It is rebuilt with the sitemap every day.
Out-of-sample check on strategy pages
2026-11-04
Strategy pages now compare the Sharpe ratio that the walk-forward parameters had on their training windows with what they earned in the following six months, and with the default parameters over the same months. They also show how many settings in the parameter grid had a positive Sharpe ratio. Across the 486 strategy-and-market pairs with parameters to tune, the median Sharpe ratio fell from 0.32 on the training windows to 0.19 out of sample; the walk-forward glossary entry has the details.
EA backtests now cover the latest twelve months
2026-11-04
All 1,063 converted EAs were re-run on four markets with hourly data from November 2025 to November 2026 (previously May 2025 to May 2026). 1,119 results are reliable enough to rank. This now happens automatically every month.
EA pages show results month by month
2026-11-03
For each market with a reliable backtest, EA pages now show the share of months with a gain and the best and worst month, with dates. EA backtests are also re-run every month on the latest twelve months of hourly data.
Strategy results now always use walk-forward optimization
2026-11-03
Until now, each strategy kept its default parameters whenever they scored better over the whole history than the walk-forward run. That choice looked at the full sample, so it flattered the results: it applied to 106 of 576 strategy-and-market results, which scored a median of 5.6 points higher than their walk-forward versions. Every result now uses walk-forward optimization, so some numbers and rankings moved down.
Identical EA backtests are no longer counted
2026-11-03
85 EAs produced exactly the same backtest as another EA on every market — for example 14 unrelated EAs with identical results — which means the automatic conversion lost what made them different. Their 298 results are now labeled "Same as other EAs", with links to the twins, and are left out of the EA Ranking, the category statistics and the blow-up risk evidence. Reliable EA results go from 1,421 to 1,123. The article on 1,063 free MT5 EAs and the glossary have been updated.
EA pages compare each EA with its category
2026-11-03
Under the backtest table, EA pages now show how each result ranks against other EAs of the same category on the same market, for example "EURUSD: better than 82% of 37". Only reliable results count, and only when at least ten EAs of the category have one.
Two new glossary terms: recovery factor and blow-up risk
2026-11-03
The glossary now has 20 terms. Recovery factor explains the number in the EA Ranking (the median EA backtest scored about zero). Blow-up risk explains broker stop-outs, how our backtests handle them and how the EA labels work: 61 of 307 gold backtests with default lot sizes ended in a stop-out, none on three currency pairs. EA pages link to both.
Blow-up risk label for every Expert Advisor
2026-11-03
Each EA now carries a blow-up risk label built from its code and our backtests: martingale sizing, grid orders, no stop loss, several positions at once, and whether the account was stopped out in our backtest. The label shows on EA pages, in the EA Arena (with a filter), in the EA Ranking and on the category pages. 93 EAs are rated high, 66 elevated. EA categories in the app now match the EA pages.
A plain leaderboard page and a faster home page
2026-11-03
The new leaderboard page lists the top 10 strategies on every market for the last 12 months and the last five years, in English and Chinese, without waiting for the app to load. The home page now shows its summary, top strategies and latest research straight away, and every static page links to the static versions of the leaderboard, strategy library, EA categories and info pages.
Readable pages for About, Methodology, FAQ and more
2026-11-03
About, Methodology, FAQ, Changelog, Privacy and Terms now have plain HTML versions (English and Chinese) that load without JavaScript and are what search engines index. Tool views such as the EA ranking and the strategy lab are no longer offered to search engines.
Eight new research articles
2026-11-03
New data-driven articles on luck-adjusted Sharpe ratios across 72 strategies, which strategies held up in 2022, 1,063 free MT5 EAs backtested, why martingale and grid EAs blow up, moving averages on bitcoin, turtle trading since 2021, MACD with an RSI filter, and how to read drawdown, the Ulcer index and rolling Sharpe. The Keltner vs Bollinger article was rewritten with ten-market results.
Browse Expert Advisors by category
2026-11-03
Each EA category — trend following, grid, martingale, scalping and more — now has its own page with a short explanation, backtest statistics for the whole group, how many EAs are rated dangerous, and every EA sorted by how many markets it made money on.
EA backtests rerun with realistic position sizes
2026-11-03
The converted EAs kept adding orders while one was still pending and treated lots as units, so positions grew far beyond the EA's lot size and most P&L figures were wrong. All 1,063 EAs were rerun on four markets with MT5 lot sizes, 1:100 leverage, one order at a time and a broker-style stop-out. Three conversion bugs that kept some EAs from ever trading were fixed. The EA Ranking now also shows the recovery factor and on how many markets each EA made money. Results that rely on a placeholder for a custom indicator are now labeled as approximations and left out of the ranking, and EA pages explain why an EA never traded when the cause is a conversion bug.
Confidence ranges for every strategy
2026-11-03
Strategy pages show a 90% bootstrap range for CAGR and maximum drawdown, plus the chance of a loss, so you can see how much a result depends on a few lucky stretches. A new glossary entry explains how it is computed.
Clearer EA backtest status and a trading-cost slider
2026-11-03
EA backtests now say why a result is missing: the EA could not be converted automatically (most cases), it held one position to the end, or it really never traded. Strategy pages have a slider that shows the annual return after 0–50 bp of trading costs.
Shareable portfolios and quick links to new tools
2026-11-03
Portfolio weights in the compare view now live in the link, so a mix of strategies can be bookmarked or shared with one click. The home page links straight to the Strategy Lab, today's signals, the heatmap, weekly reports and the glossary.
Strategy Lab: compare your own backtest
2026-11-03
Upload an equity curve (CSV) or a MetaTrader strategy tester report and see where it ranks against every arena strategy on the same market and the same trading days. Everything is computed in your browser; nothing is uploaded.
Crisis tests, luck-adjusted Sharpe and embeddable badges
2026-11-03
Strategy pages now show how each strategy did in the Q4 2018 sell-off, the COVID crash and the 2022 bear market, its returns above and below the 200-day average, CAGR after 10 bp trading costs, and the Deflated Sharpe Ratio, which corrects for testing many strategies at once. Each strategy also has a live Arena Score badge you can embed.
Chinese strategy pages and share images
2026-11-03
All 72 strategy pages and the category pages now have Chinese versions, and every strategy has a share image with its equity curve and key numbers for social media previews.
Today's signals, strategy heatmap, weekly reports and RSS
2026-11-03
New pages show the current position of every strategy on every market, an Arena Score heatmap of strategies × markets, and an automatic weekly report (market moves, signal changes, best and worst strategies). Follow along with the RSS feed at /feed.xml (Chinese: /zh/feed.xml).
Arena Score v2 and 20 new metrics
2026-11-03
The Arena Score now ranks each strategy against the others on the same market and window across five components (return quality, drawdown control, consistency, robustness, significance), so the leaderboard no longer has blocks of ties at 100 or 0, and strategies that were almost never in the market are not ranked. Strategy pages add time in market, Ulcer index, longest drawdown, VaR/CVaR, alpha, beta, capture ratios and the probabilistic Sharpe ratio, with five new glossary entries.
Every strategy, article and EA now has its own page
2026-11-03
72 strategy pages with equity curves, results on every market and a buy-and-hold comparison; strategy category hubs; research articles in English and Chinese; a 12-term metrics glossary with live examples; and 1,515 EA pages in both languages. Pages load fast and can be found by search engines.
EA rankings now only use reliable backtests
2026-11-03
Sharpe ratios are recomputed from each equity curve (the old values used yearly returns and were often meaningless). Results with no trades, run errors or flat curves are labelled and left out of the ranking, and detail pages say why.
Site-wide search and recently viewed
2026-11-03
Press / or Ctrl/Cmd+K anywhere to search strategies, EAs, research notes and pages. The home page now remembers what you opened recently (stored only in your browser).
Redesigned header, home page and footer
2026-11-03
A cleaner navigation with a More menu, icon theme and language toggles, a new home overview with live counts, quick-entry cards, ranked top strategies, a timeline changelog, back-to-top button and a reading progress bar on articles.
Faster loading and real backtest data restored
2026-11-03
Equity curves are now split per instrument and loaded on demand, and EA analysis data only loads on EA pages, cutting the initial download from tens of MB to about 3 MB. The last complete backtest snapshot (2026-08-17) is back in place of placeholder figures, and pages show the data date clearly.
Backtest Pipeline Complete: 560 Strategies Tested
2026-05-14
All 560 auto-convertible EA strategies have been backtested on EURUSD H1 data (730 days) using backtrader. Results include total trades, win rate, P&L, max drawdown, and equity curves for each EA.
Equity Curve Visualization
2026-05-14
EA detail pages now display an equity curve chart rendered with Canvas API. Charts show ~200 downsampled points with gradient fill and theme-aware styling.
Markdown Rendering for EA Analysis
2026-05-14
EA analysis and description fields now render Markdown syntax (headings, lists, code blocks, bold/italic) using marked.js v15.0.12. Both Chinese and English views support full Markdown.
Backtest Result Auto-Sync
2026-05-14
Added sync_backtest_to_site.py to automate the full cycle: generate strategy → run backtest → update index → publish site → push to repo.
Generator Overhaul: Sub-Line + Level Reference Resolution
2026-05-14
MQ5→Python generator now resolves Bollinger Band sub-lines (.lines.bot/.top/.mid), Ichimoku sub-lines (.l.tenkan/.l.kijun/.l.senkou_span_a/.l.senkou_span_b), and 30+ level reference patterns (oversold, overbought, ADX threshold, etc.). Same-base crossover guard prevents CrossOver(close, close) false signals.
EA Comparison Feature
2026-05-12
EA Arena now supports side-by-side comparison of up to 4 EAs. Select via checkboxes, compare metrics in a modal overlay, and share comparisons via URL parameters.
MQ5→Python Backtest Pipeline
2026-05-10
Implemented Scheme 2 pipeline: LLM-assisted MQ5→structured JSON→Python backtrader template. Includes extract_logic.py (GLM-5.1 analysis), generate_backtrader.py (template codegen), and run_backtest.py (execution with progress tracking).
EA Arena: 1515 EAs Analyzed
2026-05-08
Completed analysis of 1515 MQL5 Expert Advisors with bilingual CN/EN descriptions, parameter documentation, and category classification. All EAs separated into utility and trading categories.
Bilingual EA Detail Pages
2026-05-06
Each EA now has a standalone HTML detail page with full Chinese/English toggle, markdown-rendered analysis, parameter tables, and strategy classification.
Performance Optimization: Lazy-Load EA Data
2026-05-04
Reduced initial page load from 12MB to 17KB by splitting EA catalog into a lightweight index and per-EA lazy-loaded JSON files.
SEO: 1,510+ URLs in Sitemap
2026-05-02
Expanded sitemap from 15 to 1,510+ URLs covering all EA detail pages. Added OG and Twitter Card meta tags with canonical URLs.
Cleaned Monthly Resample Warnings
2026-01-27
Updated monthly resample frequency to ME in strategy scripts to avoid pandas deprecation warnings.
Added Parameter Sensitivity Heatmaps
2026-01-27
Strategy detail pages now visualize parameter grid sensitivity using Arena Score heatmaps.
Added Latest Snapshot Panel
2026-01-27
Strategy detail headers now show the most recent data point and daily benchmark/strategy move.
Added Performance Attribution
2026-01-27
Strategy detail pages now include alpha/beta attribution with contribution curves and summary metrics.
Added Auto Publish Script
2026-01-27
Added a PowerShell script to build data, publish the site, and push the public repo for scheduled automation.
Isolated AdSense on Content Routes
2026-01-20
Tool routes now force a clean reload when ad scripts are present, preventing ads from appearing on low-content screens.
Added Donchian/Turtle Breakout Research
2026-01-20
Published a long-form deep dive into Donchian channel breakouts and the Turtle rulebook.
Added Supertrend and ATR Research
2026-01-20
Published a long-form note on supertrend signals and ATR trailing stop risk control.
Added Ichimoku Cloud Research
2026-01-20
Published a long-form guide to Ichimoku cloud signals and regime filters.
Added RSI Mean Reversion Research
2026-01-19
Published a long-form RSI mean reversion strategy analysis to reach five research articles.
Added Sitemap and Robots
2026-01-19
Created sitemap.xml and robots.txt to help search crawlers discover content pages.
Scoped AdSense to Content Pages
2026-01-19
AdSense now loads only on research and policy content, with canonical metadata updates to reduce duplicate URLs.
Added Research Articles
2026-01-13
Published long-form analyses for EMA Crossover, Parabolic SAR, Volatility Breakout, and the Arena scoring methodology.
Added Research Hub and Homepage Entry Points
2026-01-13
Research index, featured article highlights, and new Learn the Method section on the homepage.
Added Policy and FAQ Pages
2026-01-13
Expanded About/Methodology plus Privacy Policy, Terms of Use, and FAQ pages.
Added 1 Mean Reversion Strategy
2026-01-13
Ultimate Oscillator added. Total strategies now 70.
Added MR-15 Ultimate Oscillator
2026-01-13
Ultimate Oscillator mean reversion using multi-horizon buying pressure.
Added 1 Mean Reversion Strategy
2026-01-13
Detrended Price Oscillator added. Total strategies now 69.
Added MR-14 Detrended Price Oscillator
2026-01-13
DPO mean reversion using a detrended price oscillator.
Added 1 Mean Reversion Strategy
2026-01-13
TRIX Reversion added. Total strategies now 68.
Added MR-13 TRIX Reversion
2026-01-13
Triple-smoothed EMA momentum extremes for mean reversion entries.
Added 3 Trend Strategies
2026-01-13
Coppock Curve, Elder Triple Screen, Vortex Indicator. Total strategies now 67.
Added TREND-18 Coppock Curve
2026-01-13
Weighted momentum curve for long-term trend timing.
Added TREND-19 Elder Triple Screen
2026-01-13
Multi-filter trend system combining MACD, pullbacks, and breakouts.
Added TREND-20 Vortex Indicator
2026-01-13
Vortex indicator crossovers for trend entries and exits.
Added 3 Trend Strategies
2026-01-13
Triple Moving Average, Linear Regression Channel, Darvas Box. Total strategies now 64.
Added TREND-13 Triple Moving Average
2026-01-13
Three moving averages confirm trend direction before entries.
Added TREND-14 Linear Regression Channel
2026-01-13
Regression channel breakout with mean-reversion exits.
Added TREND-15 Darvas Box
2026-01-13
Box breakout strategy using rolling highs/lows.
Added Risk Contribution Analysis
2026-01-13
Portfolio Builder now breaks down marginal and percentage risk contribution per strategy.
Added 5 Advanced Strategies
2026-01-13
Volatility Regime Strategy, Momentum Crash Protection, Cointegration Basket, Flight to Quality, MACD+RSI Filter. Total strategies now 61.