Quantitative trading

The research loop

The loop is short to describe and long to do well. Write a testable hypothesis. Collect a reproducible dataset and check its coverage: missing intervals, duplicate events, time zones, contract rolls and units. Run the rule with an explicit execution model and costs. Evaluate the result on a period chosen in advance that took no part in selecting the parameters. Record the outcome, including the failures, and decide what to change next. Then change one thing.

A retail trader can begin with one instrument and one hypothesis. Coding comes later; the first task is deciding what evidence would make you reject the idea.

Where the evidence usually breaks

• Costs: an illustrative gross expectancy of a few ticks can disappear once fees and slippage on entry and exit are subtracted. Convert fees with the instrument's actual tick value.
• Leakage: a level marked with tomorrow's profile, or a present-day gamma exposure substituted for the historical snapshot, lets future information into the past.
• Selection: testing many configurations and keeping the winner measures luck unless the evaluation period was separated first.
• Coverage: a chart that loads today says nothing about historical depth or about the side classification of old trades.
• Execution: a bar-based model does not reconstruct the queue, and a touched limit price is not a fill.

Quant and orderflow together

Orderflow observations become quant inputs when they are reduced to fields the data contains: delta per level as ask volume minus bid volume, diagonal imbalances at a stated ratio, volume at price, closed-bar returns. An absorption candidate, heavy selling with limited downward progress at a pre-marked level, can motivate a hypothesis; it cannot replace a record of what was known when the decision occurred. Depth-based ideas need book history, which trades alone do not provide.

In Senzoukria

The Scripts editor holds the rule in TypeScript, Python or C++, and the Auto backtest panel runs it bar by bar on the trader's own cached history, measuring coverage before the run and including costs in the result. The panel offers a walk-forward option, a parameter search and a Gauntlet of robustness checks: account survival, walk-forward with enough windows, a permutation test, a deflated Sharpe ratio that accounts for the number of trials, the probability of backtest overfitting, maximum drawdown against the account's limit, a stress test on adverse periods and a prop-firm cycle check. Quant pane indicators such as Autocorrelation, Hurst Exponent (R/S), Shannon Entropy and the z-scores are available on the chart. A backtest run keeps its trades only for the last three runs and loses them on reload, which is one more reason to keep a written research record.

Common mistakes

• Treating a profitable chart example as a strategy.
• Combining every indicator at once instead of testing one well-documented experiment.
• Reading a favourable Gauntlet label as a certification of future edge rather than a summary of a protocol on supplied data.
• Skipping the step from backtest to replay to simulated account before considering a connected one.