Skip to content
AmiBroker TA Course

Understanding Backtest Results

Part 28 taught you to make AmiBroker produce a backtest. This part is about the harder half: reading what it produced without deceiving yourself, and using the same words AmiBroker uses so that what you write down months later still means something.

That second point sounds pedantic. It is not. AmiBroker reports a row called Annual Return %; the trading literature calls the same idea CAGR. AmiBroker reports Max. system % drawdown; almost everyone writing about systems says “max drawdown”. AmiBroker reports Sharpe Ratio of trades, computed from the distribution of individual trade returns — which is a different statistic from the Sharpe ratio a fund reports, computed from periodic equity returns, and the two are not comparable. If your notes say “Sharpe 1.8” without saying which one, the note is worthless in six months.

The rule this part follows

Section titled “The rule this part follows”

Every metric name, capitalisation and definition in these six lessons is taken from AmiBroker’s own documentation. Where a statistic that trading writing treats as standard is simply absent from AmiBroker’s report, this part says so plainly instead of implying it is there under some other heading. There is no built-in Sortino ratio, no MAR ratio, no System Quality Number, no row named Expectancy, and — the one that surprises people most — no metric of any kind for how long a drawdown lasted.

Why the vocabulary problem is a research problem

Section titled “Why the vocabulary problem is a research problem”

A backtest report is not a verdict. It is a description of what a specific rule set did to a specific universe over a specific period under a specific set of cost and execution assumptions. Strip any of those away and the numbers stop meaning anything, but they keep looking authoritative, because they are printed to two decimal places in a professional-looking table.

Most of the damage done by backtesting is done at exactly this point. Not in the simulation, which is usually doing what it was told, but in the reading — in taking a number out of its context, comparing it to a number from a different context, and deciding something. The defences are unglamorous: a fixed reading order, an explicit record of the settings, and an insistence on knowing what each number is a ratio of.

What the six lessons do

Section titled “What the six lessons do”

The first lesson establishes the process — where the report lives, what is in it, the order to read it in, and what to write down every single time. The next three work through the metrics themselves in three groups: return, risk, and per-trade statistics. Each defines the metric exactly as AmiBroker defines it, names the traps in its definition, and says what question it can and cannot answer.

The fifth lesson takes the single most quoted statistic in trading marketing and shows, with arithmetic you can check line by line, two systems that share a win rate and end the year on opposite sides of zero. The sixth turns from the summary table to the shape of the equity curve, and makes an argument that runs against most people’s instinct: a suspiciously smooth curve is a warning at least as often as it is a comfort.

One test, read six ways

Section titled “One test, read six ways”

Rather than a different toy example in every lesson, this part uses a single backtest formula throughout — an ordinary long-only breakout with its universe filter, costs, delays, stop and position sizing all written into the file. You run it once on your own data, and every lesson then talks about the report you are looking at.

Any specific figure quoted in these lessons is illustrative. The numbers are chosen to make an arithmetical point clearly and are labelled as such; they are not results the course is claiming, and they are not what your run will produce. That is deliberate. A course that quotes its own backtest returns is teaching you to read backtest returns as achievements, which is precisely the habit this part exists to break.

What you need first

Section titled “What you need first”

Part 28, in full. You need to have run a portfolio backtest and to understand what SetPositionSize(), SetTradeDelays(), ApplyStop() and the commission settings do, because half of reading a report is knowing which setting produced which number.

An AmiBroker installation with a few years of end-of-day data on a few dozen symbols is enough. Nothing in this part needs the Professional edition, with one exception noted where it arises: the MAE and MFE distribution charts on the report’s charts page are labelled Professional Edition only in the settings dialog.

What comes next

Section titled “What comes next”

Part 30 catalogues the ways a backtest can be wrong before you ever read its report, and several of its lessons are best read immediately after this one — once you can read the report fluently, the mistakes become visible in it. Parts 31 to 33 then put the report under pressure: optimization, walk-forward testing and Monte Carlo simulation all consume these same metrics, and all of them will mislead you comprehensively if you have not first learned what the metrics mean.

Part29 of 36Level4 — Trading System ResearcherPages6Estimated time2.8 hours

0 / 6 lessons in this part completed