17.1 Introduction to Quantitative Methods

The 2026 CMT Program Guide places this unit in Section Eleven: Systems and Quantitative Methods, starting with Introduction to Quantitative Methods. It sits in Advanced Techniques (26% of CMT Level I—the 132-question, 2-hour sitting, 120 scored plus 12 unscored pilots). Level I wants definitions and process, not a production desk build. The same pages also cover two Advanced Techniques subdomains at that definition depth: Systematic Trading (repeatable, testable rules instead of a one-off chart story) and Principles of Risk Management (position size, reward-to-risk, invalidation, and not staking the whole book on one trade). Independent OpenExamPrep teaching for these CMT Level I topics is not a CMT Association publication and does not claim approval, review, or partnership. Independent CMT Level I practice by OpenExamPrep is at /practice/cmt.

Section 5.3 already used reward-to-risk and invalidation on a discretionary trend trade. Section 12.2 defined survivorship bias for index membership. This unit asks the next exam jobs: how a technician outlines the investment process, what quantitative analysis is, how the scientific method maps onto a test, how survivorship bias contaminates a test file, when to use trigger / filter / value rules, how signals, strategies, and models differ, and why a signal test should precede a full backtest.

Outline the investment process

Quantitative work is a loop, not a single annotation on Friday's bar.

1. Set objectives and constraints. Return goal, maximum drawdown you can live with, liquidity, tax, and mandate (long-only equity, futures, options overlay). A breakout engine that needs short sales does not belong in a long-only account.
2. Define the universe and the clock. Which markets, timeframes, and instruments are eligible—and which names existed on each historical date. Point-in-time membership belongs here, not as a cleanup after the equity curve looks good.
3. Research and specify rules. Turn an observation into coded trigger, filter, and value logic someone else could replay.
4. Construct the trade. Convert a signal into entry, invalidation, target, and position size.
5. Implement. Execute with realistic costs and fills. Do not assume you traded the signal print.
6. Measure, monitor, and revise. Compare results to the hypothesis. Change a rule only with a new test, not after one ugly week.

A discretionary technician still uses this loop when writing a playbook. A systematic technician writes it so another analyst could replay the same data and get the same signals. That repeatability is the definition of systematic trading at Level I. It is not a claim that computers must fire the orders, and it is not high-frequency microstructure.

What quantitative analysis is

Quantitative analysis uses numbers, coded rules, and historical tests to evaluate claims about price, breadth, volatility, or sentiment. The product is a statement you can falsify: after event X, the next N-bar return (or rank, or hit rate) differs from a fair comparison on this universe, and the difference survives costs.

Narrative chart reading can generate the idea. Quantitative analysis decides whether the idea is more than a good story. Level I does not require deriving a t-statistic (that skill lives in the statistics unit). It does require knowing which object you are testing—a signal, a full strategy, or a multi-piece model—and why a test can lie.

Exam trap: calling a hand-picked gallery of winning head-and-shoulders charts a quantitative study. A gallery is a story. A study has a universe, a date rule, a holding rule, and a place for the failures.

Scientific-method steps, then investing

A working outline of the scientific method for this exam:

1. Observe a regular feature of price or of an indicator.
2. Ask a precise question (does a 20-day breakout lead the next swing, or does it lag a move that already happened?).
3. State a hypothesis that could be wrong.
4. Predict a measurable outcome (sign, magnitude, or rank of forward returns).
5. Test on data that could refute the prediction, including sample that was not used to invent the rule.
6. Analyze hit rate, average payoff, and whether costs erase the edge.
7. Conclude: keep, modify, or discard—and write down what would count as a later failure.

Applied to investing, that sequence stops "I see a head-and-shoulders, so I am bearish" from skipping to a live order. A testable hypothesis might be: a completed head-and-shoulders with a close below the neckline is followed by negative 15-session forward returns on this universe after costs. If the test fails, the pattern can still be useful language for structure. It is not, by itself, a strategy.

Out-of-sample and walk-forward splits belong to later system-testing work. At Level I, remember the principle: the data that suggested the rule is a poor place to certify the rule.

How survivorship bias impacts tests

Survivorship bias is the error of studying only names that survived a filter—still listed, still in the index, still large enough—while dropping names that failed, were acquired, were demoted, or were delisted. Section 12.2 showed it on an index membership list. In a quantitative test, the same error overstates win rate, overstates compounded return, understates drawdown, and hides the breakouts that never came back.

How it shows up in a test file:

• Backtesting "buy every S&P 500 name that closes at a 52-week high" on today's membership. Companies that were in the index, broke down, and were removed never appear as losers.
• Testing a relative-strength rotation only on funds that are still listed after 15 years. Closed funds are often the weak ones.
• Reading a vendor equity curve for a "price-pattern service" that quietly dropped failed patterns.
• Building a momentum study on names that are liquid now, then pretending you could have traded the 2008 version of that list at today's spreads.

Official total-return series from a serious index provider usually reconstruct historical membership. That is the honest history of that rule set. The bias appears when you rebuild the study from the ending list. Related traps: look-ahead membership (a constituent list that would not have been known on the historical date) and missing delisting returns (the last gap to the acquisition price or toward zero).

The repair is point-in-time data: on each date, use the universe that actually existed, and keep the exit of names that left. If the stem describes a 20-year study of current members, name survivorship bias before you praise the CAGR.

Trigger rules, filter rules, and value rules

These three rule types are how a technician turns a chart idea into something a computer—or a disciplined checklist—can repeat.

A trigger is an event. A filter is a gate. A value rule is a comparison. Mixing them is normal: a value rule may define "overbought," a filter may require an uptrend, and a trigger may be the first down-close that starts a fade. Level I wants you to name which piece is which. A common trap is calling a trend filter a "signal" when it never timed an entry. Another trap is optimizing the value cutoff until the backtest sings, then forgetting that the cutoff was a free parameter.

When to use each. Use a trigger when you need a clock (a moment to act). Use a filter when the same trigger is junk in the wrong regime (long breakouts in a confirmed markdown). Use a value rule when the idea is "enough"—enough stretch, enough RSI, enough breadth thrust—not merely "a cross occurred."

Worked stack. Filter: weekly close above a rising 40-week average. Value: 14-day RSI between 50 and 70 (momentum without a blow-off). Trigger: daily close above the 20-day high. That is one systematic long setup, not three separate "systems."

Signals, strategies, and models

Keep the vocabulary strict. The Program Guide asks you to summarize these three, and stems will mix them on purpose.

• A signal is the information: the trigger fired, the score flipped, the pattern completed. It is not yet a trade. "The 50-day average crossed above the 200-day average on Tuesday" is a signal.
• A strategy is the operational wrapper: signal plus entry logic, invalidation, exits, position size, costs, and constraints (one position per sector, no trade into the print, skip the open auction). Two technicians can share a signal and still run different strategies.
• A model is the broader representation of how inputs produce outputs. It may contain several strategies, a regime detector, or a statistical mapping from breadth and volatility into a position. Calling a single moving-average cross "the model" is sloppy at exam depth unless that cross is the entire representation.

Exam trap: reporting a strategy equity curve as if it proved the signal. Sizing, stops, and skipped trades can manufacture a curve even when the event never predicted returns. That is why the next heading exists.

Why a signal test should precede a full backtest

A full backtest builds an equity curve: compounding, stops, targets, sizing, skipped trades, and often a search over parameters. It has many moving parts. Those parts can manufacture a handsome curve even when the underlying event never predicted returns.

A signal test asks a narrower question first: after the event, what happened to forward returns (or rank, or hit rate) versus a fair comparison, on a clean universe, with a simple holding rule. Fewer free parameters. Faster to reject junk. If the signal has no predictive content, you never spend the week coding pyramiding rules.

Issues with jumping to a backtest first:

• Extra parameters (trailing stops, volatility sizing, "skip the first three signals") overfit noise.
• Look-ahead: using a moving average that includes the close you pretend you traded, or a volatility estimate that uses the same bar's range.
• Selection bias: keeping the parameter set that won the last sample.
• Confusing a money-management artifact with an informational edge.
• Spending compute and hope on an idea whose event was never informative.

The benefit of the signal test is isolation. You learn whether the idea has content before you let position sizing rescue it. A Level I sketch: 80 breakout events on a point-in-time universe; hold 10 sessions; compare mean forward return to the universe's mean over the same windows. If the breakouts are indistinguishable from the universe before costs, a gorgeous fully sized backtest is not evidence—it is decoration.

After a signal survives, then wrap it: costs, slippage, invalidation, size, portfolio constraints, and a check that five correlated breakouts are not five independent bets. That full backtest is still not live trading. Live trading adds missed fills, halted names, and the technician's willingness to follow the invalidation that was easy to write in a spreadsheet.

Level I risk-management principles inside the process

Systematic methods without risk rules are a signal generator. Fold these definitions into every strategy, at the same depth Section 5.3 used for trend-following.

Invalidation. The price or condition that means the setup is wrong. For a breakout long, that might be a close back inside the range, not an arbitrary 8% round number. The protective stop belongs at invalidation. A systematic rule that "uses a 2% stop" without tying 2% to structure has replaced invalidation with a pain threshold.

Reward-to-risk. Planned reward divided by the amount to invalidation. Example: buy 50.00, invalidation 48.50 (risk $1.50), measured target 54.00 (reward $4.00). Reward-to-risk = 4.00 / 1.50 ≈ 2.67:1. A setup that risks $2 to make $1 can still sit inside a high-win-rate process, but Level I trend teaching treats an adequate ratio against real structure as a first filter, not an afterthought. Fantasy targets in empty sky do not count.

Position size. Size so that a stop at invalidation costs a planned fraction of equity—commonly discussed in the 0.5%–2% range per idea—not so that the notional looks "small." On $80,000 equity and 1% risk ($800), a $1.50 stop allows about 533 shares. Notional is 533 × $50 ≈ $26,650; risk is still about $800 if the stop is honored. Gaps can blow through the stop. That is why size still cannot be allowed to threaten the account even when the spreadsheet risk looks tidy.

Do not risk the book on one trade. Five correlated breakouts in the same sector are one bet. Doubling down to "get even" converts a planned loss into account risk. A systematic process pre-commits size and invalidation so a hot streak cannot silently raise leverage. "The book" here is the trading account and, for a professional, the book of client capital—not a slogan about the CMT charter. Risk of ruin is what happens when one idea, or one cluster of look-alike ideas, is large enough that a normal losing streak ends the program.

Worked loop. Signal test says 20-day breakouts have a modest positive 10-session edge on a point-in-time large-cap universe. Strategy wrapper: enter next open, invalidation = close back through the breakout level, target = 2.5 times the risk, size = 1% of equity at the stop, max two names per GICS sector, skip if reward-to-risk against the next resistance is below 1.5:1 after costs. That is systematic trading plus risk management. A gallery of five perfect charts is neither.

Exam habits for this unit

When a stem shows a beautiful equity curve, ask whether a signal test happened first and whether the universe is point-in-time. When a stem mixes a moving-average cross, an ADX gate, and an RSI cutoff, label trigger, filter, and value. When a stem asks why the backtest came before the signal study, say too many free parameters and money management can fake an edge. When a stem gives entry, stop, and target, compute R:R and size from the stop, then check whether several names are the same bet.

Key Takeaways

• Investment process: objectives, universe, rules, trade construction, implementation, review
• Quantitative analysis falsifies a coded claim; a chart gallery is not a test
• Survivorship bias in tests = studying only who is still there
• Trigger fires, filter gates, value compares; signal ≠ strategy ≠ model
• Signal test first; size from invalidation; do not risk the book on one idea