You download a return series called Momentum. Have you acquired a stock-selection strategy? The file contains monthly returns, but no individual stock fields, eligibility decisions, or order quantities. It can help explain a portfolio without generating your account's trades.
The missing connections run through inputs, eligibility, ranking, holdings, and execution. Start with five fictional stocks and produce a checkable signal table. Then examine why the top-ranked selection can lose money and what a published factor-return series actually provides.
What you will be able to do
- Distinguish time-series questions from cross-sectional ones, and characteristics, signals, factor returns, and risk exposures from one another.
- Calculate a signal from supplied fields while preserving its timing, eligibility, and missing-data reasons.
- Save a T17 factor specification with a mechanism, alternative explanations, implementation limits, and unfinished real-data requirements.
The C7-82 table is a standalone synthetic preparation exercise. It is not C9's complete real-data package. Historical membership, corporate actions, and stock-level fields remain requirements for actual factor research.
Compare one asset over time—or several assets now
C2-A compares one ETF's state with its own history. That is a time-series question. A cross-sectional question compares several objects at the same decision time, such as which has the highest return over a defined earlier interval. Both may use the word momentum without sharing a rule or evidence.
A company characteristic describes an object: price relative to earnings, financial resilience, size, or past return. A signal applies a specified calculation to inputs to produce a ranking or prediction variable. Having a research use does not establish predictive ability.
A factor return is a return series produced by particular portfolio groups and weights. A risk exposure in a model describes how a portfolio relates to movements in a shared return source and needs further estimation. A stock characteristic alone does not supply an exposure coefficient or alpha.
CFA Institute's strategy overview connects factor-based research with data processing, evaluation, portfolio construction, and trading costs.[1] Follow those connections instead of treating a list of factor names as a completed study.
Translate the name into fields
Value might use price relative to earnings, book equity, or cash flow. Quality might concern profitability, stability, or financial strength. Size requires a market-capitalization definition; low volatility needs a return frequency, window, and estimator. “Buy cheap, high-quality small companies” is not yet reproducible.
Names also vary across studies. French's five-factor definitions construct HML from specified value and growth portfolios and RMW from operating-profitability portfolios.[2] They do not define every low-P/E screen or every list labeled good companies.
Choose one main price signal here to reduce simultaneous changes. C9's candidate baseline decides after decision month m closes. Using the total-return series available then, divide P(m−1) by P(m−12) and subtract one. Each P is a value from that stock's consistently defined total-return index, not an interchangeable raw price without dividend or split treatment.
This skips the most recent decision month, m. The endpoints m−12 and m−1 are eleven months apart. French's monthly momentum documentation likewise specifies its own historical window and portfolio construction; this simple long-only exercise does not reproduce that academic portfolio.[3]
Put dates around the notation. If m is December, the numerator is the end of November and the denominator is the previous December's month-end. The current December contributes no return to the feature. After the December-close decision, execution still starts in the following month. Skipping a month in the feature and waiting for an executable time are two distinct rules.
Calculate five rows before revealing their outcomes
All assets, history, and trading conditions below are synthetic. The C7-82 materials separate signal inputs from later returns. The exercise supplies traceable identities and historical eligibility at m. A, B, C, and E have the other required history; D lacks its m−1 endpoint. Values are each stock's index points, not dollar quotes.
Freeze selection first. Exclude objects missing required fields from this decision's eligible pool, rank signals from highest to lowest, and select the top 20%, rounding the number upward. Resolve trading ties alphabetically by stable asset ID. Use only this signal. The 20% is a teaching parameter, not an optimized recommendation.
| ID | P(m−12) | P(m−1) | Signal | Treatment |
|---|---|---|---|---|
| A | 100 | 110 | 10% | Eligible |
| B | 100 | 120 | 20% | Eligible |
| C | 50 | 60 | 20% | Eligible |
| D | 100 | Missing | Unavailable | Exclude and retain reason |
| E | 100 | 90 | −10% | Eligible |
A gives 110/100 minus one, or 10%. C's level is only 60, yet 60/50 minus one gives 20%, tying B. Comparing index levels across stocks would answer the wrong question. Filling D with a zero return would turn unavailable information into an invented observation.
Four stocks qualify. Four times 20% is 0.8, rounded upward to one. The order is B, C, A, E, so select B. Allocating this exercise's strategy capital entirely to the selected group would leave a single concentrated holding. That is a property of this tiny example, not household-allocation or live-position advice.
Keep D in the original table and exclusion record. Eligibility can change at the next decision. No valid signal this month does not mean the company never existed.
Match selection to the allowed holding interval
The decision requires month m's completed close. It cannot support an assumed purchase before that close. Here, orders may be attempted at the first normal trading day's open in m+1; the next rebalance point is the corresponding open in m+2. For this main exercise, sufficient fills are assumed.
Now reveal the synthetic next-holding-period gross total returns, including distributions: A 4%, B −2%, C 6%, and E 0%. They span those two opening times, with no outside cash flows. A month-end-to-month-end return series would be a different label.
The selected B earns −2% before costs. A benchmark allocating one quarter to each eligible stock at the start, without interim rebalancing, earns the average: 4 minus 2 plus 6 plus 0, divided by four, or 2%. Selection therefore trails that benchmark by 4 percentage points.
This neither disproves momentum nor permits selecting C after observing its 6% gain. It checks whether the frozen rule was followed. Preserve the unfavorable result. G83 examines association, grouping, and time evidence without replacing this table with a more flattering one.
A changed list still needs a transaction
If B later loses purchase eligibility and C becomes the new target, distinguish admission to a new position from treatment of an existing holding. Selling B still needs availability, price, and cost information. Missing terminal values or a suspension do not authorize deleting it from the account.
State a turnover convention. Here, add actual sell and buy amounts and divide by whole-account equity immediately before rebalancing. A half-turnover convention takes half that ratio. Risky-asset value and whole-account equity are different denominators.
Use a separate cost snapshot: $10 cash plus old B worth $200 gives $210 equity. Sell B for $200, then buy C for $200. Both fill completely; each commission is 5 basis points of its traded value. Assume no price movement, slippage, or tax in this snapshot.
Each order costs $0.10, so final cash is $9.80, C is worth $200, and equity is $209.80. Combined traded value is $400; dividing by $210 gives about 190.476% turnover, or 95.238% when halved. This is commission arithmetic, not a complete net C9 backtest.
C9's baseline 5/5-basis-point costs also include directional slippage and defined account policies, which need matching prices in a full ledger. SEC investor guidance explains how costs reduce investment results.[4] The main gross-return table cannot quietly become a net account by borrowing this separate snapshot.
What a published factor file cannot fill in
French's momentum factor compares stronger and weaker historical-return groups using its size divisions, weights, and other rules.[3] Its returns describe that research construction. They are not all the original fields and constraints required by your stock account.
B's long-only −2%, the pool's equal-weight 2%, and a published long/short momentum return are different objects. Subtracting the first two gives active return, not risk-adjusted alpha. Industry, market, size, and other common exposures still need investigation.
A published series can support interpretation or a benchmark. It cannot supply absent historical membership, contemporary prices, corporate actions, or data permission. Downloading it does not establish independent construction and validation of a stock-selection strategy.
Financial signals add another clock
Price signals need corporate-action and historical-version checks. Financial signals also separate the reporting period from publication. Profit for a quarter ending March 31, released in May, cannot enter an early-April historical ranking. Later revisions cannot be moved backward either.
The financial-factor route adds G19, G20, G22–G24, and G70: sources, business, statements, cash flow, ratios, and definitions. A price-to-earnings measure needs compatible price dates, shares, and earnings definitions. Dividing historical prices by today's latest database field would not supply them.
Record a mechanism and alternatives for every signal. Earlier relative strength could reflect persistence, common risk, reversal, or crowding. A historical relationship might also be a winner selected from many attempts. A factor name cannot settle these explanations.
Diagnose two errors yourself
Explanation question: Why do C and B have the same signal despite C's lower index level?
Application question: An analyst assigns D a zero return, then gives C priority after seeing its next-period 6%. Which rules changed? Save your answers before checking.
Use each stock's own endpoints: 120/100 and 60/50 both equal 1.2. Cross-stock point levels are not directly comparable prices. D's missing value became an invented known zero; C's new priority changed a frozen tie policy after outcome inspection. Both alter the study version.
Supply one genuinely available observation
In a separate variation, D's m−1 value was legally obtained and verified as 130 before the original decision. Other inputs and rules stay fixed. Recalculate eligible count, selection count, and first choice. What if 130 was entered only in m+1?
If genuinely available beforehand, D's signal is 30%. Five stocks qualify, 20% selects one, and D ranks first. No subsequent return for D is supplied, so its investment performance cannot be calculated.
If the field appeared only in m+1 without evidence of earlier availability, the original decision retains the missing value. Later discovery is not proof that the workflow could have used it earlier. Establish availability before treating a rebuilt table as historical knowledge.
Save your T17 specification and signal table
Record mechanism, universe and identities, observation window, raw fields, total-return definition, formula, direction, availability, missing values, ties, selection count, holding and rebalance times, weights, benchmark, and cost version. Retain inputs, signal, eligibility, and reason for every row, not only selected names.
Hand-calculate A and C first, then ask a tool to produce all five rows under the frozen rules. Compare the first discrepancy. Mark a tool that could not execute or export, or was not actually run. A correct synthetic mechanism table does not establish real C9 data acquisition.
For real inputs, separately retain permissions, historical identities and membership, corporate actions, dates, gaps, data versions, and every attempt. Can another reader rebuild the signal from raw fields? Does measured return start at the allowed execution time? Are unfinished exposure and data checks still visible?
Build the connection before evaluating the result
Identify whether the comparison follows one object over time or several objects at once. Define fields, windows, direction, and eligibility. Missing values, ties, and availability belong to the rule; none can be chosen afterward to improve results.
Then connect signal, holdings, turnover, and cost, using a consistent benchmark. Separate active return, factor return, and model alpha. A checkable table starts research; cross-period evidence, real inputs, and implementation still require their own work.
References
- CFA Institute: Active Equity Investing—Strategies. Public 2026 overview and summary, not the paid curriculum.
- Kenneth R. French: Description of Fama/French 5 Factors. Specific research constructions, including HML and RMW.
- Kenneth R. French: Monthly Momentum Factor. Historical window and high-minus-low portfolios; this exercise is not a replication.
- Investor.gov: How Fees and Expenses Affect Your Investment Portfolio. July 23, 2025. The book's basis-point inputs are teaching assumptions, not broker quotes.
Sources checked October 30, 2026. C7-82 inputs, returns, and cost snapshot are synthetic; real C9 data and effectiveness remain unverified. AI-assisted draft; independent AI full-text and bilingual review is complete. Human financial review and target-reader testing have not been conducted and are not scheduled for this production round.