Skip to content
AmiBroker TA Course

Developing a Complete Trading Research Process

By this point in the course you can validate a database, define a universe, write an exploration, rank a market cross-sectionally, build a strategy, backtest it at portfolio level, find the bias that made it look good, optimise without deceiving yourself, and size a position so that a bad run does not end the experiment. Each of those was taught as a separate skill, because that is the only way to teach them.

They are not separate in use. On any given evening they run in a fixed order, each one consuming what the previous one produced, and the whole sequence has to finish inside the time you actually have. This part is about that sequence: what it contains, what each stage hands to the next, where the human belongs in it, and how you find out afterwards whether the process or the person was the problem.

What this part assumes you already have

Section titled “What this part assumes you already have”

Not much software, but a certain amount of prior work. The four lessons here refer constantly to things built earlier — a validated database from Part 2, an Analysis workflow from Part 12, a ranking tool from Part 13, a backtested strategy from Parts 27 to 29, a risk model from Part 34. If you have been reading rather than building, the lessons still make sense, but the exercises will not have anything to run on.

What you do need is a decision about scope. A research process for one instrument on a weekly chart and a research process for two thousand instruments on a daily chart are the same shape and completely different in cost. Pick the smaller one first.

The claim this part makes

Section titled “The claim this part makes”

A trading process fails in one of four places, and they are not equally likely:

  • The data was wrong, and everything downstream inherited it.
  • The rules were wrong, in the ordinary sense that the evidence never supported them.
  • The rules were fine and were not followed, which is a different problem with a different fix.
  • Nothing was wrong, and a run of losses inside the range the evidence already predicted was mistaken for a signal to change something.

Most published material about trading systems addresses only the second of those. This part insists on separating all four, because the corrective action for each is different and applying the wrong one is how a defensible system gets dismantled by its own owner.

Where this part stops, and why that is deliberate

Section titled “Where this part stops, and why that is deliberate”

The pipeline in Lesson 1 ends at a journalled decision and a periodic review. It does not continue into order routing, and no lesson in this course teaches automated live execution.

That is a choice, and Lesson 3 argues for it at length rather than asserting it. The short version: the moment a formula can place an order, every defect in it changes category. A mislabelled array that produced a wrong exploration row now produces a position. The skills that make that safe — operational monitoring, reconciliation, failure handling, the regulatory obligations that apply to you specifically — are not the research skills this course teaches, and pretending otherwise would be the least honest thing in it.

AmiBroker does have an automated-trading interface for one broker, and the course says so plainly, with its documented status and its own author’s safety warnings, so that you know what exists and can go and study it properly. What the course will not do is show you a formula that sends an order and leave you to discover the rest.

How to read the four lessons

Section titled “How to read the four lessons”

In order, and preferably with an Analysis window open. Lesson 1 gives the map; Lessons 2, 3 and 4 each take one stretch of it and make it concrete. Every lesson ships a complete AFL formula, and the four formulas are one set: a data-validation gate, a universe and regime definition, a candidate review sheet, and a journal export. Run in sequence they are a working, if minimal, research process.

The last lesson is the one to read twice. Separating strategy performance from your own execution performance is the single measurement that tells you which of the four failure modes above you are actually looking at, and almost nobody does it.

Part35 of 36Level4 — Trading System ResearcherPages4Estimated time2.0 hours

0 / 4 lessons in this part completed