SYSTEMATIC INVESTING · DAILY ETF RESEARCH

Evidence before conviction.

Inspect the data. Challenge the signal. Understand the risk.

↓ Export research record

01 / OUT-OF-SAMPLE

Does the signal survive the costs?

StrategyWindow

Growth of $10,000

Selected strategy SPY buy & hold

Drawdown from peak

Initial capital included

ROBUSTNESS

Cost sensitivity

Total one-way trading cost. Same signals, changed execution assumptions.

UNCERTAINTY

A Sharpe ratio is an estimate.

95% interval · 21-session block bootstrap · 500 resamples. Full out-of-sample period, zero cash rate. No multiple-testing correction.

Same dates. Same cost assumptions.

Strategy	CAGR	Sharpe	Max drawdown	Turnover / year

Turnover = sum of absolute asset trades / pre-trade equity; a complete rotation counts as 2×. Equal-weight rebalances daily. SPY pays its initial entry cost.

02 / PERSONAL RESEARCH

Make the exposure tangible.

Paper allocation from the latest completed session. No brokerage connection.

Hypothetical portfolio ($)

Next-session target

Rebalance worksheet

Enter current dollar holdings to estimate the difference to the research target. Values stay in this page.

Fractional dollars, no share rounding. Estimated costs are illustrative; taxes, bid/ask conditions and settlement are not modeled. Research strategies cap each ETF at 30% and do not borrow.

03 / MODEL ACCOUNTABILITY

Every prediction has a past.

Ridge regression · no opaque model

Only available information

21/63/126-session returns, 21-session volatility and distance to the 200-day mean. All features end at the decision close.

Chronological selection

Three expanding training folds, a two-session gap and train-only scaling. Select ridge α from 1, 10 and 100 by validation MSE.

Walk forward, then repeat

Refit every 63 sessions using matured targets. Hyperparameter search stops before the terminal evaluation window; weights never use a future label.

Training provenance

Asset

Prediction starts	Last training feature	Last label available	Observations	Ridge α	Window

Showing the latest 8 fits for the selected asset. Full fold metadata and validation scores are in the exported research record.

Out-of-sample forecast quality

All matured predictions. Compare MSE with a zero-return forecast and direction accuracy with always predicting up.

Asset	OOS R² vs zero	Rank IC	Direction accuracy	Always up

04 / DATA INTEGRITY

Nothing disappears into a clean dataset.

Event type

Date	Asset	Event	Reason / disposition

05 / RESEARCH CONTRACT

Know what this experiment can tell you.

The execution clock

Session t closes → features and target weights become known.
Session t+1 closes → simulated fill, commissions and slippage.
Session t+2 closes → first return earned by the new target.

Daily total-return prices incorporate splits and distributions once. Trading costs are self-financing and include weight drift. No missing bars are forward-filled.