Website of Vivek Kumar Malviya · Not the official website of Angel One Ltd.Never share OTP or PIN · Stay safe →
Vivek Malviya.Authorised Person · Angel One Ltd

Backtesting without fooling yourself

Treat the backtest as an experiment with assumptions, not a promise.

Build a reproducible experiment

Specify the universe, time range, data availability, signal time, execution model, costs, position sizing and corporate-action handling. Include instruments that disappeared when the research question requires a historical universe.

Look-ahead bias uses future information. Survivorship bias selects assets using later survival. Overfitting selects rules that exploit sample-specific noise. These problems are different and need separate checks.

A bar hides the sequence

A fictional bar opens 100, trades between 90 and 110, and closes 105. Both a 95 stop reference and 108 objective fall within its range. OHLC alone does not reveal which occurred first. Choosing the favourable sequence systematically can inflate results.

Keep evaluation independent

Develop on one sample, validate the process and preserve an untouched evaluation period. If you repeatedly retune to a holdout, it is no longer independent. Keep failed experiments and search history. A positive historical result still depends on future conditions and actual execution.

Diagram: process at a glance

Fix the hypothesis → Model fills and costs → Test unseen data → Report uncertainty

Check your understanding

Can a backtest assume the target occurred first whenever both levels lie inside a bar?

No. Use finer observations or an explicit defensible convention; do not select the favourable sequence after seeing outcomes.

खाता खोलें →WhatsApp →
ACCOUNT OPENING HELP

आपकी शुरुआत यहाँ से होगी।

यह केवल डिज़ाइन का नमूना है। आपका मोबाइल नंबर कहीं भेजा या सेव नहीं किया गया। असली प्रक्रिया देखने के लिए नीचे मौजूदा वेबसाइट खोल सकते हैं।

असली प्रक्रिया समझें →
FIND YOUR WAY

आप क्या ढूँढ रहे हैं?

Search articles, guides and account help.