AI-powered intraday trade signal generator for Indian stock markets, built on the Kronos financial foundation model — accepted at AAAI 2026.
⚠️ Disclaimer: This tool is for research and educational purposes only. It does not constitute financial advice. Past model performance does not guarantee future results. Always use a hard stop-loss order with your broker. Trade only what you can afford to lose.
Current experiment:
- Model: Kronos-base
- Market Universe: NSE Large Cap + Mid Cap + Small Cap
- Interval: 1 minute
- Prediction Horizon: 1 trading day
- Signal Tracking: Enabled
The built-in tracker is collecting live signal outcomes. Performance statistics will be published once a statistically meaningful sample size has been accumulated.
Kronos is a decoder-only foundation model pre-trained specifically on financial candlestick (K-line) data — OHLCV sequences — from over 45 global exchanges, covering more than 12 billion K-line records. Unlike general-purpose time series models, Kronos is designed from the ground up for the unique noise characteristics of financial markets. It uses a specialized tokenizer that discretizes continuous price and volume data into discrete tokens, then applies autoregressive pre-training to learn temporal and cross-asset patterns.
In zero-shot benchmarks, Kronos outperforms the leading time series foundation model by 93% on price forecasting RankIC and achieves 9% lower MAE on volatility forecasting.
This project wraps Kronos into a complete signal generation and tracking system for the Indian stock market (NSE/BSE). Every evening after market close, it:
- Scans NSE for today's top gainers and losers across Large Cap (Nifty 100), Mid Cap (Nifty Midcap 150), and Small Cap (Nifty Smallcap 250)
- Analyses trends using RSI, ADX, SMA20/50, weekly/monthly momentum, RVOL, and OBV
- Analyses sentiment using FinBERT on recent Google News headlines — BULLISH / BEARISH / NEUTRAL per stock
- Predicts price using Kronos — fed the last 512 candles of OHLCV history at your chosen interval
- Generates signals with entry, a Kronos-native target (the model's own predicted high/low — no fixed cap), a stop-loss clamped to a 1.0–2.5% risk band, and R:R ratio — only when Kronos and the trend agree
- Tracks outcomes over time to measure real-world accuracy against actual market data
NSE Archive CSVs → Large / Mid / Small cap universe
yfinance OHLCV → Historical candles (1h / 15m / 5m / 1m)
Trend Analyzer → RSI · ADX · SMA · RVOL · OBV · Momentum
Google News RSS + FinBERT → Sentiment per stock (BULLISH / BEARISH / NEUTRAL)
↓
Kronos Foundation Model → Predicts next N trading days (OHLCV)
↓
Signal Generator → LONG / SHORT / NO TRADE
Entry · Target · Stop-Loss · R:R · Confluence · Sentiment
↓
Prediction Tracker → SQLite log · WIN/LOSS evaluation · Performance report
| Requirement | Version |
|---|---|
| Python | 3.10+ |
| CUDA (recommended) | 12.8+ |
| GPU VRAM (recommended) | 4 GB+ |
GPU strongly recommended: CPU inference is possible but very slow — Kronos-base may take several minutes per stock on CPU.
Note on
--interval 1m: Kronos has a 512-candle context window. One full NSE session is 375 one-minute candles, so--interval 1mgives Kronos only approximately 1.5 trading days of history as context — very limited. For broader market context use--interval 15minstead, which fits approximately 20 trading days within the same 512-candle window.
git clone https://github.com/shiyu-coder/Kronos.git
git clone https://github.com/jvaidya10/kronos-india.git
cd kronos-indiapip install -r requirements.txt
pip install einops==0.8.1 matplotlib==3.9.3# For CUDA 13.2 (RTX 40/50 series)
pip3 install torch torchvision --index-url https://download.pytorch.org/whl/cu132
# For CUDA 12.8
pip3 install torch torchvision --index-url https://download.pytorch.org/whl/cu128# Windows
setx KRONOS_PATH "C:\path\to\Kronos"
# Linux / Mac
export KRONOS_PATH="/path/to/Kronos"python -c "import torch; print('GPU:', torch.cuda.get_device_name(0))"pip install streamlit
streamlit run app.pyOpens at http://localhost:8501 — all arguments as dropdowns, per-step progress bars, styled scanner tables (green/red), trend analysis table, colour-coded signals table (green for LONG, red for SHORT), and tracker controls built in.
# Daily scan — all cap tiers, predict next 3 days, log signals
python main.py --track
# Intraday only — 15m candles, predict tomorrow, large cap
python main.py --interval 15m --days 1 --cap large --track
# Test on specific stocks — by symbol or company name
python main.py --symbols RELIANCE TCS INFY --samples 5
python main.py --symbols "reliance industries" "hdfc bank" "bajaj finance"| Argument | Default | Description |
|---|---|---|
--variant |
small |
Kronos model: mini (4.1M) · small (24.7M) · base (102.3M) |
--cap |
all |
Cap tier: large · mid · small · all |
--top |
10 |
Top N gainers + losers per tier |
--interval |
1h |
Candle size: 1h · 15m · 5m · 1m |
--days |
3 |
Trading days to predict (1 day = 7/25/75 candles by interval) |
--samples |
20 |
Kronos ensemble samples — more = stable but slower |
--symbols |
— | Skip scanner, predict specific symbols or company names |
--save |
off | Save actionable signals to outputs/ CSV |
--track |
off | Log signals to tracker DB for outcome evaluation |
See USAGE.md for full documentation with examples.
| Interval | Candles/day | Context window | History | Best for |
|---|---|---|---|---|
1h |
7 | ~73 trading days | 730 days | Swing / multi-day |
15m |
25 | ~20 trading days | 60 days | Intraday (recommended) |
5m |
75 | ~7 trading days | 60 days | Detailed intraday |
1m |
375 | ~1.5 trading days | 7 days | Scalping (limited) |
================================================================
NSE INTRADAY SIGNAL PIPELINE — Powered by Kronos
Model: Kronos-small | Interval: 15m | Predicting 1d (25 candles)
================================================================
[1/5] Scanning NSE — top 10 gainers & losers: LARGE, MID, SMALL
TOP GAINERS :: Large Cap (Nifty 100)
symbol ltp change_pct change
TECHM 1543.20 4.00 59.30
INFY 1202.50 3.58 41.60
[3/6] Analysing weekly & monthly trends...
TECHM Monthly:BULLISH (+6.6%) Weekly:BULLISH (+7.7%)
RSI:77.28 ADX:14.30 SMA20=above SMA50=above
RVOL=1.77x[SPIKE] OBV=RISING => STRONGLY BULLISH [score +7]
[4/6] Analysing news sentiment...
TECHM Sentiment: BULLISH (0.79, 6 headlines)
[5/6] Running Kronos predictions (samples=20)...
Predicting: TECHM...
======================================================================
ACTIONABLE TRADE SIGNALS (Next 1 Trading Day)
Target: Kronos range (min 1.5%) | Max SL 2.5% | Min R:R 2.0:1
======================================================================
[BUY ^] TECHM [HIGH confidence] Confluence: STRONG ***
Entry: 1543.20
Target: 1638.88 (+6.2%)
Stop Loss: 1504.62 (-2.5%)
R:R Ratio: 2.48:1
Trend: STRONGLY BULLISH (score +7/8)
Sentiment: BULLISH (0.79, 6 headlines)
* Kronos predicts upside 6.2% (target 6.2% / stop 2.5%)
* Monthly: BULLISH (+6.6%) | Weekly: BULLISH (+7.7%)
* RSI 77.28 | ADX 14.3 | Score +7/8
* RVOL 1.77x [VOLUME SPIKE — confirms move] | OBV RISING
Why the target isn't a fixed percentage: Kronos forecasts a full price range, so the target is its predicted high (long) or low (short) and the stop is its predicted opposite extreme, clamped to a 1.0–2.5% risk band. A fixed 5–7% target both missed real 2–4% large-cap moves (filtered out as NO TRADE) and clipped the occasional 8–10% move. Letting Kronos's own range drive the levels also makes the 2:1 reward:risk filter meaningful. See USAGE.md for the full rationale.
Each stock gets a score from -8 to +8 based on:
| Factor | Bullish | Bearish |
|---|---|---|
| Monthly momentum (>2%) | +1 | -1 |
| Weekly momentum (>1%) | +1 | -1 |
| Price above SMA 20 | +1 | -1 |
| Price above SMA 50 | +1 | -1 |
| RSI > 55 / < 45 | +1 | -1 |
| Volume spike in trend direction | +1 | -1 |
| OBV rising / falling | +1 | -1 |
| ADX > 25 (trend strength bonus) | +1 | -1 |
| Level | Score | Meaning |
|---|---|---|
STRONG *** |
≥ +3 / ≤ -3 | Trend strongly agrees with Kronos |
MODERATE ** |
+1 or +2 / -1 or -2 | Trend leans same way |
WEAK * |
0 | Trend is neutral — take smaller position |
AGAINST TREND |
Opposite sign to signal | Trade blocked |
Every signal logged with --track is evaluated after the prediction window closes. The tracker walks through candles in order to determine which level — target or stop-loss — was hit first, making evaluation more realistic than simply checking the end-of-period close.
- The candle interval used for a run is stored alongside each signal and used when fetching actual price data during evaluation — a 15m signal is evaluated on 15m candles, not 1h
- eval_by date is calculated using NSE holiday-aware business days — weekends and official NSE holidays are both skipped
- Evaluation uses the actual next-day open price as the realistic entry point, not the signal's logged close price — targets and stop-losses are recalculated from there
- The EXPIRED exit price is the last close of the eval_by date specifically, not any earlier candle
# Log signals after each run
python main.py --track
# Evaluate outcomes after prediction window closes
python tracker.py evaluate
# View full performance report
python tracker.py report
# List all logged signals
python tracker.py show
# Force-evaluate overdue signals skipped due to data gaps
python tracker.py evaluate --force
# Import a saved CSV into the tracker (for runs done without --track)
python tracker.py import --csv outputs/signals_YYYYMMDD_HHMM_<variant>.csvTotal signals evaluated : 24
Wins : 15 (62.5%)
Losses : 7
Expired (no hit) : 2
Avg win P&L : +5.8%
Avg loss P&L : -2.4%
Expectancy : +2.7% per trade
--- By Confluence Level ---
STRONG : 8 trades | Win rate 75% | Avg P&L +3.8%
MODERATE : 12 trades | Win rate 58% | Avg P&L +2.1%
--- By Confidence Level ---
HIGH : 10 trades | Win rate 80% | Avg P&L +4.1%
MEDIUM : 11 trades | Win rate 55% | Avg P&L +1.8%
LOW : 3 trades | Win rate 33% | Avg P&L -0.6%
--- By Sentiment ---
BULLISH : 9 trades | Win rate 78% | Avg P&L +3.9%
NEUTRAL : 11 trades | Win rate 55% | Avg P&L +1.4%
BEARISH : 4 trades | Win rate 25% | Avg P&L -1.2%
--- By Cap Tier ---
LARGE cap : 10 trades | Win rate 70% | Avg P&L +3.2%
SMALL cap : 8 trades | Win rate 50% | Avg P&L +0.8%
--- By Direction ---
LONG : 18 trades | Win rate 67% | Avg P&L +2.9%
SHORT : 6 trades | Win rate 50% | Avg P&L +1.1%
--- By Interval ---
1h : 14 trades | Win rate 64% | Avg P&L +2.4%
15m : 10 trades | Win rate 60% | Avg P&L +3.1%
--- Streaks ---
Current streak : 3 × WIN
Best win streak : 5 in a row
Worst loss run : 3 in a row
kronos-india/
├── main.py # CLI entry point — orchestrates the full pipeline
├── app.py # Streamlit web dashboard
├── tracker.py # Signal logger + outcome evaluator + report
├── requirements.txt # Python dependencies
├── USAGE.md # Full argument reference
├── pipeline/ # Internal pipeline modules
│ ├── market_scanner.py # NSE gainers/losers by cap tier
│ ├── data_fetcher.py # Historical OHLCV via yfinance (multi-interval)
│ ├── trend_analyzer.py # RSI, ADX, SMA, RVOL, OBV, momentum scoring
│ ├── sentiment_analyzer.py# FinBERT news sentiment (Google News RSS)
│ ├── predictor.py # Kronos model wrapper (GPU-accelerated)
│ ├── signal_generator.py # LONG/SHORT signal with entry/target/SL
│ ├── symbol_resolver.py # Company name -> NSE symbol lookup (fuzzy search)
│ └── symbol_names.csv # 2,456 NSE equities: symbol + company name
├── tests/
│ └── test_tracker.py # Unit tests for outcome evaluation and business day logic
├── outputs/
│ ├── signals_*.csv # Saved signal CSVs (--save)
│ └── tracker.db # SQLite prediction log (--track)
└── ../Kronos/ # Kronos repo (cloned separately)
| Model | Parameters | Context | Speed (RTX 5070) |
|---|---|---|---|
Kronos-mini |
4.1M | 2048 candles | ~1-2 sec/stock |
Kronos-small |
24.7M | 512 candles | ~2-5 sec/stock |
Kronos-base |
102.3M | 512 candles | ~8-12 sec/stock |
Models are downloaded automatically from HuggingFace Hub on first run (~500MB for small) and cached locally. All subsequent runs are fully offline.
Unit tests cover the tracker's outcome evaluation logic and NSE holiday-aware business day calculation — the two functions that directly affect whether performance claims can be trusted.
python -m pytest tests/- NSE gainers/losers by market cap tier (Large/Mid/Small)
- Multi-interval candles (1h / 15m / 5m / 1m)
- Trend confluence scoring (RSI, ADX, SMA, RVOL, OBV)
- GPU-accelerated Kronos inference (CUDA)
- Prediction tracker with WIN/LOSS evaluation
- Tracker: interval-aware evaluation, NSE holiday scheduling, actual-entry P&L, streak tracking
- Web dashboard (Streamlit) with live progress, styled tables, tracker controls
- FinBERT news sentiment analysis (Google News RSS, no API key required)
- Fine-tuning Kronos on Indian market data
- Live candle loop (re-predict every candle during market hours)
- Telegram / email alerts for actionable signals
Built on Kronos by Yu Shi, Zongliang Fu, Shuo Chen, Bohan Zhao, Wei Xu, Changshui Zhang, and Jian Li — accepted at AAAI 2026.
@misc{shi2025kronos,
title={Kronos: A Foundation Model for the Language of Financial Markets},
author={Yu Shi and Zongliang Fu and Shuo Chen and Bohan Zhao and Wei Xu and Changshui Zhang and Jian Li},
year={2025},
eprint={2508.02739},
archivePrefix={arXiv},
primaryClass={q-fin.ST},
url={https://arxiv.org/abs/2508.02739}
}