SwingTrading

This is not financial advice. Use this software entirely at your own risk. Trading can result in substantial financial loss, and market data, technical signals and AI-generated research can all be incomplete or incorrect. Always test a strategy thoroughly with a paper-trading account before considering its use with a live account and real money.

Quickest route: run it inside Codex

The quickest way to use this project is to ask Codex to clone the repository, install its dependencies, scan the current company universe supplied in data/us_uk_large_mid_mega_cap.csv with a 12-day short and 24-day medium moving average, and then analyse the resulting BUY candidates fundamentally. This lets Codex perform the installation, technical screen and structured fundamental review as one workflow instead of requiring you to step through every stage individually.

You can give Codex this instruction:

Clone https://github.com/acnash/SwingTrading.git, install the required dependencies, run the companies in data/us_uk_large_mid_mega_cap.csv using a 12-day short and 24-day medium moving-average crossover, then feed the BUY results through three separate, independent fundamental and recent-news reviews, then use render_ai_review.py to calculate a three-review consensus mean score. Summarise the consensus classifications and distinguish actionable entries from signals waiting for a pullback or further confirmation.

Codex should inspect the commands and generated evidence before drawing a conclusion. The resulting classifications are research outputs rather than trade instructions, and every candidate should be independently verified.

SwingTrading is a research-oriented moving-average crossover screener for the S&P 500, FTSE 100, and optional custom watchlists. It uses completed daily candles from Yahoo Finance and produces deterministic technical signals for subsequent fundamental and news review.

This software is for research and screening. It does not execute trades and does not provide investment advice. Market data can be delayed, incomplete, or adjusted. Verify every signal independently before making a decision.

Default rule:

• BUY: 20-day SMA crosses above 50-day SMA.
• SELL: 20-day SMA crosses below 50-day SMA.

Example: Premier Foods 12/24 BUY signal

This Premier Foods (PFD.L) daily chart demonstrates the shorter crossover configuration used in the example full-market run. It displays the 12-day SMA, 24-day SMA and 200-day SMA together with RSI(14) and daily volume. The 12-day average crossed above the 24-day average, generating a technical BUY signal, while the entry filters found moderate RSI and limited price extension.

The subsequent fundamental review awarded Premier Foods an 80% evidence-support score, classified as STRONGLY SUPPORTS BUY SIGNAL. That percentage measures how strongly the available fundamentals, guidance, recent news and supplied technical context support the crossover. It is not a forecast or an expected investment return.

Install

python -m venv .venv
pip install -r requirements.txt

Activate the environment first with source .venv/bin/activate on macOS/Linux or .venv\Scripts\Activate.ps1 in Windows PowerShell.

Run

python stock_signal_screener.py

Outputs are written to output/:

• all_signals.csv
• buy_signals.csv
• actionable_buy_signals.csv
• wait_for_pullback_signals.csv
• wait_for_confirmation_signals.csv
• wait_for_volume_confirmation_signals.csv
• sell_signals.csv
• failed_symbols.csv
• buy_signal_review_prompt_1.txt
• buy_signal_review_prompt_2.txt
• buy_signal_review_prompt_3.txt

failed_symbols.csv records unavailable symbols and insufficient histories so a partial market-data response cannot pass silently.

Run each numbered prompt in a separate Codex task so that every reviewer reaches its conclusion independently. Save their JSON arrays as ai_review_1.json, ai_review_2.json and ai_review_3.json, then aggregate them with:

python render_ai_review.py ai_review_1.json ai_review_2.json ai_review_3.json \
  --output-csv fundamental_consensus.csv

The renderer requires exactly three files, checks that they contain the same tickers, and returns a three-review consensus mean score together with the score range. It derives the final classification from the rounded consensus mean. BUY CANDIDATE requires a mean score of at least 65%, WATCH / NO TRADE covers 50% to 64%, and a score below 50% produces REJECT BUY / REVIEW EXIT. These labels support research decisions and remain subject to the technical entry status, independent verification and risk controls.

Change the moving-average windows

For example, 10-day / 30-day:

python stock_signal_screener.py --short 10 --medium 30

Responsible market-data access

Daily histories come from Yahoo Finance through the unofficial yfinance package. The defaults deliberately trade speed for reliability and lower request pressure:

• 25 symbols per batch
• no more than four yfinance worker threads
• a random two-to-four-second pause between live batches
• exponential backoff and three retries for failed symbols only
• a persistent .cache/market_data cache with a 12-hour freshness window

An ordinary second run within 12 hours uses the local cache. Force a new fetch only when required:

python stock_signal_screener.py --refresh-cache

The controls are configurable, for example:

python stock_signal_screener.py --batch-size 20 --threads 2 \
  --min-pause 3 --max-pause 6 --max-retries 4

Avoid scheduling overlapping runs, because separate processes do not share a rate limiter.

Late-entry protection

Every raw BUY crossover is assigned an entry status. The default safeguards are:

• WAIT_FOR_PULLBACK when RSI(14) is at least 68, price is at least 4% above the short SMA, the five-day gain is at least 8%, or the latest close is within 1% of the 52-week high
• WAIT_FOR_CONFIRMATION when price is within 2% of its prior 60-session high and fewer than two of the last three sessions traded at 1.2 times their respective 20-day average volumes
• ACTIONABLE_BUY when neither condition applies

Volume Flow Indicator confirmation

Every signal now includes a deterministic 0-to-100 VFI BUY confirmation index. The implementation uses a 130-session VFI, 30-session volatility threshold, 2.5 times average-volume cap, 0.2 coefficient and five-session EMA signal line. The index awards points for positive VFI, position above its signal line, a recent bullish signal-line cross, positive five-session direction, improvement over 20 sessions and proximity to its 20-session high.

• 80-100: STRONG_ACCUMULATION
• 65-79: SUPPORTS_BUY
• 50-64: MIXED_EARLY_ACCUMULATION
• 35-49: WEAK_VOLUME_SUPPORT
• 0-34: DISTRIBUTION_OR_ABSENT_SUPPORT

An otherwise actionable crossover with an available VFI index below 50 becomes WAIT_FOR_VOLUME_CONFIRMATION. VFI remains a confirmation layer alongside the 12/24 crossover and the three-review fundamental consensus mean; the scores are kept separate so technical and fundamental weaknesses remain visible.

These filters do not rewrite the original 20/50 crossover. They separate trend detection from entry timing. Every threshold has a corresponding command-line option, including --max-rsi, --max-short-extension-pct, --max-five-day-gain-pct, --max-52-week-high-distance-pct, and the resistance and volume-confirmation options.

52-week peak protection

The screener calculates the high and low from the adjusted daily highs and lows of the latest 252 completed sessions. It also reports the closing price's percentage distance below the 52-week high and its position within that range. By default, a raw BUY whose latest close is no more than 1% below the 52-week high is retained for audit purposes but assigned WAIT_FOR_PULLBACK, regardless of volume confirmation. This prevents a crossover at the top of its annual range from becoming an actionable entry.

The lookback and threshold can be changed with --high-low-lookback and --max-52-week-high-distance-pct. If 252 completed sessions are unavailable, the 52-week measures are recorded as unknown and do not delay the signal.

200-day trend confirmation

The 200-day SMA is currently a confirmation factor rather than a hard rejection rule. A valid short/medium crossover can therefore remain a raw BUY signal when its closing price is below the 200-day SMA, allowing the screener to identify a possible early recovery. In the generated AI-review rubric, a close above the 200-day SMA earns five technical-support points; a close below it earns zero.

Candidates below the 200-day SMA should normally remain on the watchlist until a completed daily candle closes back above that average, preferably with improving volume. The screener does not currently assign a dedicated WAIT_FOR_200_SMA_RECLAIM status, so this additional decision is made during the fundamental and technical review stage.

Add your own shares

Create a CSV:

ticker,company,market
AAPL,Apple Inc.,US
AZN.L,AstraZeneca PLC,UK

Then:

python stock_signal_screener.py --custom-csv my_stocks.csv

You can use --no-sp500 and/or --no-ftse100 to disable the built-in universes.

Bundled US and UK large/mid-cap universe

The repository includes data/us_uk_large_mid_mega_cap.csv, containing S&P 500, FTSE 100, and FTSE 250 constituents in the screener's required ticker,company,market format. These indices provide a liquid large-, mega-, and mid-cap universe while excluding penny-stock and micro-cap universes by construction.

Run only the bundled list with:

python stock_signal_screener.py --no-sp500 --no-ftse100 \
  --custom-csv data/us_uk_large_mid_mega_cap.csv

The committed file was generated on 14 August 2026. Index membership changes, so regenerate it from the current public constituent tables when needed:

python tools/update_universe.py

Tests

pip install -r requirements-dev.txt
python -m pytest

The tests exercise symbol normalisation and BUY/SELL crossover classification without downloading market data.

Important

Run after the relevant market has closed if you want the newest daily candle to represent a completed session. The "Evidence support %" generated by the AI rubric is not a probability of profit or a recommendation to trade.

The constituents are retrieved from public web tables at runtime, so upstream page changes can temporarily break universe discovery. Yahoo Finance access is provided through the unofficial yfinance client and is subject to its own availability and terms.