Technical, Fundamental and Systematic Analysis

People argue about technical versus fundamental analysis as though they were rival football teams. They are not competing answers. They are different questions, asked of the same market, using different evidence, on different time horizons. A third approach — systematic or quantitative analysis — is not a question at all but a method, and it can be applied to either of the other two.

By the end of this lesson you should be able to state what technical analysis claims in a form specific enough to be wrong, summarise what a fundamental analyst is actually doing, separate the choice of data from the choice of decision process, and say precisely where this course sits.

Three questions, one market

• Technical analysis asks: given what price and volume have done, what has tended to happen next, how often, and by how much?
• Fundamental analysis asks: what is this claim on future cash flows or resources worth, and how does that compare with the price being asked for it?
• Systematic analysis asks: what does the evidence say when the question is put to a whole universe of instruments, over a defined period, using rules stated in advance?

The third is a method rather than a source of data, which is why it sits awkwardly in the usual two-way argument. You can be systematic about price data or about balance sheets. You can be unsystematic about either.

What technical analysis actually claims

The working definition this course uses is deliberately plain:

Technical analysis is the analysis of market behaviour using price, volume and related market-generated information.

AmiBroker’s own Technical Analysis Guide puts it more traditionally, as the examination of past price movements in order to forecast future price movements, and notes that it applies to any instrument whose price is influenced by supply and demand.

That word forecast is where this course parts company with the traditional phrasing, and it is worth being explicit about why. A forecast is a statement about what will happen. Nothing in a price series supports statements of that kind. What a price series can support is a statement about a distribution: in situations resembling this one, over this universe and this period, outcomes were distributed like this, with this much spread, based on this many observations. That is a far weaker claim, and it is the only kind this course is willing to make.

?Reality check

The strong version of the technical claim — that the chart contains everything worth knowing, because the price already reflects all available information — is stated so broadly that no observation could count against it. Whatever happens next, the claim absorbs it. A statement that nothing could refute is not a finding about markets; it is a position about markets. The narrower version below can be tested, which is why the course works with that one instead.

Here is the testable version, and it is worth reading twice because every reality check in this course is an instance of it:

Conditional on a state defined from past price and volume, the distribution of returns over the next N bars differs from the unconditional distribution over the same universe and period — by enough, and consistently enough, to survive realistic costs.

Every element of that sentence is doing work. Conditional on a state defined from past data rules out anything that peeks forward. The distribution rather than the average keeps dispersion in view. Differs from the unconditional distribution forces a comparison against the base rate, which the next lesson makes the central discipline. Over the same universe and period prevents the comparison from being rigged. And survive realistic costs is where a great many otherwise interesting effects quietly die.

What technical analysis is genuinely good at, stated without inflation:

• Objective state definition. “Above the 200-day average” is unambiguous and computable across ten thousand symbols. Very little in fundamental analysis is that crisp.
• Measurement of what has happened. Prices, ranges, volumes and their relationships are facts, and they are cheap and available.
• Risk control. Structure in price gives you somewhere defensible to place a stop and something to size a position against. Parts 5 and 34 build on this.
• Comparison at scale. Ranking a thousand instruments against each other on a common measure is routine. Parts 12 and 13 do this repeatedly.
• Repeatability. A rule expressed in price terms can be applied identically today and in 2004, which is what makes historical evaluation possible at all.

And the boundary around all of it: the transaction record contains transactions, and nothing else. That boundary, and the several other constraints that come with it, are the whole subject of the next lesson.

Fundamental analysis in one page

The fundamental analyst studies the thing itself rather than the record of its trading.

For a company: revenue and how it is earned, margins and whether they are defensible, the capital the business consumes, the balance sheet, the quality of the accounts, competitive position, management incentives — and then the price being asked. Common methods include discounting expected future cash flows at a rate that reflects both time and risk, comparing valuation multiples against similar companies, and estimating what the assets alone are worth. For a commodity: production, inventories, consumption and the cost of the marginal producer. For a currency or a bond: interest rates, inflation, policy and credit.

Two honest observations about it:

Its output is a range, not a number. Change the growth assumption slightly and a discounted cash flow valuation moves substantially. Practitioners know this; presentations of a single “target price” usually conceal it.

It says very little about timing. An instrument can trade far from any defensible estimate of value for years. That is not a defect of the method — it is answering a different question — but it is the reason many otherwise fundamental investors take an interest in price behaviour.

There is also a serious data problem if you want to test fundamental ideas systematically: you need point-in-time data, meaning the figures as they were reported and known on the day, not as they were later restated. Databases that quietly contain restated figures produce backtests that could not have been executed. It is the same class of error as the survivorship problem Part 2 covers, and it is expensive to fix properly.

AmiBroker itself can store and use a set of fundamental data fields, viewable in its Information window and reachable from a formula, but this course does not depend on them. The reason is practical: the price and volume path can be followed completely with freely available end-of-day data, and the fundamental path cannot.

Quantitative and systematic analysis

Being systematic means: state a hypothesis; express it as rules a computer can apply without judgement; run it over a defined universe and period; measure the outcome against a benchmark; keep some data back; decide according to a criterion you wrote down beforehand.

Its strengths are exactly the weaknesses of informal analysis. Claims become checkable. Assumptions — about costs, timing, universe and data — must be stated explicitly, because the computer will not supply them for you. Results can be reproduced by someone else, or by you in a year, which is when most people discover what they actually did.

Its characteristic failure is equally specific: the harder you search, the better your best result looks, even when there is nothing there. Try two hundred variations and the best of them will look impressive on the data you tried them on. This is not a minor caveat; it is the dominant risk in systematic research, and Parts 30 to 32 are largely about defending against it.

Discretionary versus systematic is a separate axis

Here is the distinction that clears up most of the confusion. The data you use and the way you decide are two independent choices.

	Discretionary decision	Rule-based decision
Price and volume data	A chartist reading structure and judging each case on its merits	A backtested trend-following system executed as written
Company and economic data	A traditional value investor building a case for each holding	A factor screen rebalanced on a fixed schedule

All four cells are populated by serious practitioners. What differs between the columns is not quality but what can be evaluated, and how.

A rule-based process can be evaluated on history — with every bias trap Parts 30 to 33 describe, but it can be done. A discretionary process cannot be evaluated that way, because the historical record contains the prices but not the judgements. The only honest evaluation of a discretionary process is a journal of decisions recorded before the outcomes were known, accumulated over enough decisions to mean something. Part 35 sets that up properly.

The practical consequence is an accounting rule rather than a prohibition: every discretionary element you add to a tested system makes your historical evidence less applicable to what you will actually do. Adding one may still be the right call. Pretending the backtest still describes the result is not.

How the three combine in practice

Most working processes use all three, in distinct roles:

• Fundamental or structural criteria define the universe. Which market, which kinds of company, what size, what index membership. Combined with the liquidity filter from lesson 3, this determines what you will even consider.
• Price-based rules handle timing and risk. Entry, exit, stop placement, position size, and the market regime gate that decides whether to be active at all.
• The systematic method decides whether any of it is worth doing. It is the referee, not a third player.

Two disciplines make the combination work. Decide in advance what happens when the inputs disagree — a favourable valuation with deteriorating price behaviour, or the reverse — because deciding in the moment is how a rule quietly becomes a preference. And keep the roles separate in your records, so that when results disappoint you can tell which component was responsible.

This course sits in the top-right cell of the table, using price and volume with rules and evidence, with AmiBroker as the instrument for expressing the rules and measuring the results. It is not a claim that the other cells are empty. It is a claim that this is the cell where a reader can check the work.

Recap

You can now state the technical claim in a form specific enough to be tested, and distinguish it from the untestable version that circulates more widely. You know what fundamental analysis is trying to establish and why timing is not its strength. You can separate the choice of data from the choice of decision process, and place any approach you meet into one of four cells. And you know the accounting rule: discretion is permitted, but it is subtracted from the evidence.

The final lesson of this part turns the testable version of the claim into a working method, and introduces the loop that the remaining thirty-six parts run over and over.

Check your understanding

Question 1. Which of these is a testable version of the technical claim?

• The chart reflects everything worth knowing about an instrument
• Price action reveals the intentions of large participants
• Conditional on a state defined from past price and volume, the distribution of the next N bars’ returns differs from the unconditional distribution over the same universe and period, by enough to survive costs
• Markets move in repeating patterns driven by human psychology

Show the answer and why

Answer: Conditional on a state defined from past price and volume, the distribution of the next N bars’ returns differs from the unconditional distribution over the same universe and period, by enough to survive costs

Only the third names the condition, the outcome, the horizon, the comparison and the universe, so a measurement could contradict it. The others are broad enough to accommodate any observation, which makes them positions rather than findings.

Question 2. A factor screen ranks companies on reported earnings figures and rebalances every quarter with no human input. Where does it belong?

• Discretionary, because someone chose the factor
• Technical, because it is run on a computer
• Rule-based decision applied to company data
• It cannot be classified

Show the answer and why

Answer: Rule-based decision applied to company data

The data source and the decision process are independent axes. This process uses company data with rules, so it occupies the bottom-right cell. Choosing the factor in the first place is research, not discretion in execution — although how many factors were tried before that one was chosen is a question Part 30 takes very seriously.

Question 3. Why does testing a fundamental idea over history require point-in-time data?

• Because older data is less accurate
• Because reported figures are frequently restated later, so a database of current figures contains numbers nobody knew on the day
• Because exchanges only publish fundamentals annually
• Because currency values change over time

Show the answer and why

Answer: Because reported figures are frequently restated later, so a database of current figures contains numbers nobody knew on the day

Using restated figures lets the test act on information that did not exist at the time, which is a look-ahead error. It belongs to the same family as survivorship bias, and it is one reason systematic fundamental research is far more expensive to do honestly than price-based research.

Question 4. You add a rule permitting yourself to skip signals that "look wrong" on the chart. What has this done to your backtest?

• Improved it, because judgement is being applied
• Nothing, as long as the skipping is rare
• Weakened its applicability, because the historical results describe a process you are no longer following
• Invalidated the underlying data

Show the answer and why

Answer: Weakened its applicability, because the historical results describe a process you are no longer following

The backtest measures the rules as written. Any discretionary override means the tested process and the executed process differ, and the size of the difference is unknown. That may still be a sensible thing to do, but the evidence no longer covers it, and the only way to evaluate the override is to record those decisions prospectively.

Check my answersClear