End-to-End Strategy Development Workflow

A typical flow from raw data to live execution. This pairs naturally with a coding agent (e.g. Claude Code) for automated parameter exploration and strategy generation.

Prerequisite

This page assumes you have installed alpha-forge (binary) per Getting Started and have run alpha-forge system init in your working directory (so forge.yaml and data/ exist). All commands invoke alpha-forge directly from the working directory.

If you're working inside the alpha-trade developer monorepo, prepend FORGE_CONFIG=forge.yaml uv --directory alpha-forge run alpha-forge ... to each command.

1. Fetch historical data

Save historical OHLCV data for a target symbol locally.

alpha-forge data fetch 'USDJPY=X'

Symbol naming for FX / Futures / Crypto

yfinance (the default provider) requires fixed suffixes per asset class.

Asset class	Examples
US stocks / ETFs	SPY, AAPL, QQQ
FX	USDJPY=X, EURUSD=X, GBPJPY=X (always =X)
Futures	CL=F (WTI), GC=F (Gold), ES=F (S&P)
Crypto	BTC-USD, ETH-USD (hyphen)

Quote symbols containing shell metachars (e.g., 'USDJPY=X').

2. Create a strategy from a template

Generate a JSON scaffold, edit parameters, and register it.

How to list available templates (F-005)

There is currently no dedicated alpha-forge strategy template list command. Use one of the following to discover template IDs.

1. Trigger the error message (fastest) — pass an unknown template name and the error prints the full list of available templates:

   $ alpha-forge strategy create --template _unknown_ --out /tmp/dummy.json
   ❌ Unknown template name: _unknown_. Available templates:
     sma_crossover_v1, rsi_reversion_v1, macd_crossover_v1,
     bbands_breakout_v1, grid_bot_template, hmm_bb_pipeline_v1,
     donchian_turtle_v1
   

2. Check the documentation — each template's details (indicator stack, target markets, recommended use cases) are catalogued in Strategy Templates.

alpha-forge strategy create --template sma_crossover_v1 \
  --out data/strategies/usdjpy_sma_v1.json

Three fields you should always edit in the JSON

The generated JSON inherits the template's defaults, so edit at minimum:

1. strategy_id: leaving it as sma_crossover_v1 collides with the built-in template. Change it to a unique value (e.g. usdjpy_sma_v1).
2. name: a human-readable label.
3. target_symbols: defaults to []. Either set the symbol list (e.g. ["USDJPY=X"]) or pass it on each alpha-forge backtest run <SYMBOL>.

If you plan to optimize, also fill in optimizer_config.param_ranges. (It works even when null, but explicit ranges are easier to reproduce.)

Then save it to the strategy DB.

alpha-forge strategy save data/strategies/usdjpy_sma_v1.json

Skip DB registration with --strategy-file

alpha-forge backtest run accepts --strategy-file <path>, which loads JSON directly without DB registration — handy during rapid iteration.

3. Run a backtest

Validate the strategy against historical data.

alpha-forge backtest run 'USDJPY=X' --strategy usdjpy_sma_v1

# Show the chart URL and open it in your browser
alpha-forge backtest chart usdjpy_sma_v1 --open

Visualize results in your browser (alpha-visualizer)

The 📊 View charts via alpha-vis serve line printed at the end of alpha-forge backtest run points at the separate OSS package alpha-visualizer. Run alpha-vis serve and your browser opens the Equity / Drawdown / trades / metrics view (install instructions).

4. Optimize parameters

Bayesian search with Optuna (TPE), then apply the best result.

alpha-forge optimize run 'USDJPY=X' --strategy usdjpy_sma_v1 \
  --metric sharpe_ratio --trials 300 --save

# Apply the saved result file (optimize_usdjpy_sma_v1_<timestamp>.json) as a new strategy
# Pass the base ID (no suffix) to --to-strategy. The CLI appends _optimized automatically, registering usdjpy_sma_v1_optimized
alpha-forge optimize apply data/results/optimize_usdjpy_sma_v1_<timestamp>.json \
  --to-strategy usdjpy_sma_v1

When the best score is -inf

Every trial returned NaN. Common causes: the optimization range is too narrow, or the period has too few trades. Re-check optimizer_config.param_ranges and widen the data range.

Visualize optimization results in your browser (alpha-visualizer)

Optimization runs saved with --save are rendered as Grid heatmaps / sensitivity plots / Top-trial equity curves under alpha-visualizer's Optimize view. Start alpha-vis serve in your working directory (e.g., quickstart/).

5. Walk-forward validation

Detect overfitting with out-of-sample testing.

What is a Walk-Forward Test (WFT)? (F-006)

Running alpha-forge optimize run alone optimizes parameters across the entire period, which often produces a "curve-fitted" strategy that overfits the very data it was tuned on. WFT cures this by splitting the period into equal-sized windows and, for each window, optimizing on the In-Sample (IS) portion and then scoring on the unseen Out-of-Sample (OOS) portion. If OOS performance stays close to IS performance, the strategy is more likely to be robust across time.

Term	Meaning
IS (In-Sample)	Training period — the first half of each window, used by Optuna for optimization
OOS (Out-of-Sample)	Test period — the second half of each window, scored with the optimized params
Window	One equal-sized partition. --windows 5 splits the full period into 5
IS/OOS pair	The IS score and OOS score for each window

Rule of thumb: if OOS Sharpe is at least half of IS Sharpe, the strategy leans robust. A high IS that collapses on OOS suggests curve fitting. See alpha-forge optimize walk-forward CLI reference for the full option list.

alpha-forge optimize walk-forward 'USDJPY=X' \
  --strategy usdjpy_sma_v1_optimized --windows 5

# Sensitivity analysis (point at the optimization result JSON file)
alpha-forge optimize sensitivity data/results/optimize_usdjpy_sma_v1_<timestamp>.json

When all WFT windows show OOS 0 trades

Short data periods leave each window without any trades. For FX / 1d, aim for ~5 years (~1,250 rows). Either fetch a longer history (alpha-forge data fetch '<SYM>' --period 5y) or coarsen with --windows 2.

Visualize WFT results in your browser (alpha-visualizer)

alpha-visualizer's Optimize view renders the WFO composite equity curve (IS / OOS spliced) plus an IS-vs-OOS stability heatmap. Launch with alpha-vis serve.

6. Generate Pine Script

Export a TradingView alert script from the optimized strategy.

alpha-forge pine generate --strategy usdjpy_sma_v1_optimized

Output: output/pinescript/usdjpy_sma_v1_optimized.pine

Related commands

See CLI Reference for the complete option lists. Next step: Bringing Pine Scripts into TradingView.

See actual output samples

For output formats, equity curve examples, optimization results, and Pine Script samples, see Output Examples.