Alpha Signal from Market Data

The scenario

A research lead drops a daily price series on your desk and says: "find something in it." Your job is to engineer a predictive feature, turn it into a tradable signal, and — crucially — backtest it honestly: no lookahead, real risk-adjusted numbers, and an out-of-sample check to see if the edge is actually there.

Where this shows up

Task brief

# Alpha Signal from Market Data

**Role relevance:** The canonical quantitative-research take-home: find a signal, backtest it honestly.
**Estimated time:** 60-90 minutes
**Difficulty:** Advanced
**Format:** Jupyter notebook (.ipynb) + daily price CSV; requires numpy, pandas, matplotlib

## What you are given
- `market_data.csv` - a daily price series
- `alpha_signal_starter.ipynb` - data loading + a Sharpe helper; the signal is yours

## What you must deliver
1. Returns and a predictive feature (no lookahead)
2. A signal turned into daily PnL
3. Annualised Sharpe, hit rate, turnover, cumulative return
4. An in-sample / out-of-sample check

## Constraints
- The position must use only information available before the return it earns.

## Submission note
Build the signal in the notebook, run the evaluation, then read the mark scheme.

Your tasks

1. 01Load the daily series, compute returns, and engineer at least one predictive feature (e.g. short-term reversal or momentum).
2. 02Turn the feature into a position with no lookahead, and compute the strategy's daily PnL.
3. 03Evaluate it: annualised Sharpe, hit rate, turnover and the cumulative return.
4. 04Split into in-sample and out-of-sample windows and report whether the edge survives.

How you're assessed

The full points-based mark scheme is included with the pack.

What you'll learn

• ▸How to turn a feature into a position and measure the strategy's risk-adjusted return.
• ▸Why lookahead bias quietly inflates backtests — and how to avoid it.
• ▸Why an out-of-sample check is the difference between a signal and an overfit.

Study alongside