“Buy when price strengthens; sell when it weakens.” Two tools receive that instruction. One buys today, the other tomorrow, and their return lines diverge. Which tool made an error? Before choosing the better curve, ask what “strengthens” means and when the required information becomes available.
Quantitative research connects numerical information, mathematical or statistical methods, and decision rules in a process someone can repeat from the same inputs. Software can calculate and carry out steps. The researcher still needs to explain the question, evidence, and limits of the result.
What you will be able to do
- Distinguish data, characteristics, signals, strategies, portfolios, and execution.
- Reconcile a small account and explain why observing a signal differs from completing a transaction.
- Turn a strategy idea into a flow diagram with rules, materials, and unresolved evaluation questions.
Make the same instruction mean the same decision
Strength might mean a rise today, a price above a past average, or stronger performance than other stocks. Each definition asks a different question and requires different data. Without a definition, a tool may select defaults that produce precise results for a strategy you never intended to investigate.
The rules also need an investment universe, observation time, target quantity, execution method, and exception policy. Buying one share, twenty shares, or an entire account produces different holdings. Fees, missing prices, and an inability to sell can change the account as much as the original signal does.
A person following explicit rules manually can use a systematic method. Software automatically repeating an ambiguous or incorrect process does not establish effectiveness. CFA Institute's backtesting overview connects hypotheses, rules, portfolio formation, rebalancing, and evaluation in one process. We will first shrink that chain to something you can check by hand. [1]
Five days without additional market complications
Use the book's synthetic C4-S1-v1 account, not actual ETF prices or the long-history C2 strategy. The five days are successive trading days. Start with $1,000 cash and no shares, with no external flows, dividends, splits, taxes, or cash interest. Settlement delays are not modeled.
After each close, target twenty shares if the closing price is strictly above $11; otherwise target zero. Do not trade when holdings already equal the target. Adjust at the next listed day's open at the supplied price, paying $1 per successfully filled order. If cash is insufficient, reject the entire order without a fee. There is no leverage or fractional trading. Slippage is zero solely to simplify this exercise.
| Trading day | Open, dollars | Close, dollars |
|---|---|---|
| D1 | 10 | 10 |
| D2 | 10 | 12 |
| D3 | 13 | 11 |
| D4 | 10 | 10 |
| D5 | 10 | 10 |
At D1's close, 10 is not above 11, so the account remains at zero shares and $1,000 cash. D2's close of 12 changes the target to twenty shares. That signal is known after the close. You cannot go back and buy at that day's $10 open or assume an ordinary order already filled at $12.
At D3's $13 open, the exercise supplies a full fill. Twenty shares cost $260 plus the $1 fee, leaving $739 cash. At the $11 close, the holding is worth $220. Account equity is $739 plus $220, or $959. The $41 reduction from the starting value combines $40 of price loss after purchase and $1 of purchase cost.
D3's close equals 11. The strict inequality therefore sets a zero target, producing an order to sell twenty shares the next day. D4's $10 open produces $200 proceeds less a $1 fee, taking cash from $739 to $938. Neither D4 nor D5's close triggers another purchase. The final account value is $938.
Check the result through another route. Twenty shares bought at 13 and sold at 10 lose $60; two fees add $2. Subtract $62 from $1,000 to recover $938. The daily-equity route and the completed-trade route should reconcile. A disagreement calls for finding omitted cash or price changes, not averaging the two answers.
The return is negative 6.2%. It establishes what these five synthetic inputs produce under these rules, without showing that the rule generally gains or loses money. The exercise checks timing, cash, and holdings before investigating an economic strategy's usefulness.
Six different jobs inside the account
Data are the supplied daily opens and closes. Real research also needs dates, security identity, units, corporate actions, and provenance. A column of numbers appearing on screen is not enough to establish research-ready inputs.
A characteristic, or feature, describes an object using data. This minimum example takes the closing price directly; other tasks might calculate past returns or an average. A signal applies a condition, such as greater than 11. Keep the measured description separate from whether a decision condition is met, rather than combining both as “AI likes the stock.”
A strategy adds holding, adjustment, and exit rules to the signal. Here, targets are twenty or zero, with next-open execution, fees, and an insufficient-cash policy. A portfolio governs how assets jointly use capital. One security plus cash still requires checking whether the target is affordable.
Execution turns a target into fills and remaining orders. The ledger then updates cash, shares, and equity. A target of twenty shares does not establish twenty shares actually held. Missing quotes, halts, or inadequate quantity require a recorded actual state rather than an ideal fill invented for the strategy.
The sequence is data, features, signals, targets, attempted execution, account updates, and evaluation. You can delegate steps to AI or an application while retaining their inputs, outputs, and checks. Not writing code does not remove the need to understand these connections.
Accuracy can rise while the account loses
Some quantitative work uses prediction models, but prediction accuracy and investment results measure different things. In a separate synthetic example, ten equal-size trades include eight gains of $1 and two losses of $6, without fees. Eighty percent win, yet $8 minus $12 leaves a $4 loss.
Even if predicted direction corresponds to each trade's gain or loss here, the larger losses outweigh the high success rate. Costs, unequal positions, and execution constraints can add further differences. Evaluate returns, drawdowns, expenses, and the capital path instead of substituting one classification score for an account record.
A profitable observation does not establish a reliable relationship either. The rule might fit the selected interval by chance, or many trials may have been reduced to one displayed winner. SEC investor education distinguishes hypothetical backtested results from actual performance. Historical prices do not turn simulated orders into real fills. [2]
Start with a research map, not an algorithm ranking
Trend questions examine whether a change persists, often turning prior performance into a holding rule. Mean-reversion questions examine whether a defined deviation closes, facing the possibility that it continues instead. Both can use simple methods and both can fail. Their names do not establish a ranking of sophistication or returns.
Cross-sectional selection compares multiple objects at the same time, requiring the eligible universe and information then available. Event research focuses on announcements or other events, making publication and execution times essential. Neither can substitute today's surviving companies or later knowledge for historical choices.
Locate work by question, signal, holding period, and data needs. Intraday, daily, and monthly tasks differ in data and implementation demands. Transparent, low-frequency exercises with few objects offer a manageable starting point, while still carrying market risk and model error. Low frequency is not a synonym for safety.
Separate the economic idea from the prediction algorithm. Trend or value questions can use different algorithms. Regression, trees, and deep learning are methods for describing or predicting relationships. A text model extracting announcements or suggesting hypotheses is not, on its own, a fully specified investment strategy.
A GPU is also a resource for particular workloads. It supplies neither trustworthy data nor a research mechanism or cost advantage automatically. Define the task and check a small example before the later math and resource guides. A calculator is sufficient for this five-day ledger.
Fundamental analysis can inform explicit rules
Understanding businesses and using data can work together. Investigate an economic mechanism, then define observable conditions for a screening rule or further inquiry. Metric definitions, publication dates, and sample scope must be clear. Values reported later cannot enter earlier historical choices simply because they are convenient now.
Some judgments remain difficult to quantify reliably. Retain a qualitative note or narrow the question when consistent data are missing. Asking AI to fill every company's missing value merely to produce a score would hide the limitation. Rules should clarify inputs and boundaries, rather than force every question into a ranking.
C2 has a separate complete specification for low-frequency ETF-and-cash research, requiring real prices, corporate actions, calendars, costs, and benchmark materials. It differs from the fixed $11 threshold here. This chapter supplies no complete real-data packet or backtest ledger for C2; passing the synthetic account is not validation of that strategy.
Recalculate two steps yourself
Why can D2's observed $12 close not become a same-day $12 fill in this ledger? Then diagnose a D3 closing-equity record of $999 justified as “only the purchase fee was deducted.” Give correct cash and equity before checking the answer.
The rule forms its signal after the close and attempts the next trading day's open, supplied as $13. Purchase cost is $261, leaving cash of $739. Closing shares are worth $220, giving $959 equity. Omitting the $40 price loss is not an alternative fee convention; it leaves part of the position's change unaccounted for.
Change the price, not the strategy
Keep all rules and other inputs unchanged, but change D3's opening price from 13 to 12. Fees remain $1 per successful fill. Calculate cash after purchase, equity at D3's close, and cash after D4's sale. Explain why this is not independent evidence from another strategy test.
Purchase cost is $240 plus $1, leaving $759. D3's shares still close at $220, so equity is $979. D4's net proceeds are $199, leaving $958. The extra $20 equals a $1 lower purchase price across twenty shares. This is an input-perturbation accounting check, not investment evidence from new real observations.
Save a rule flow and a gaps table
Turn “buy strength” into your own small task. Specify the object, data, feature, observation time, signal, targets, execution time, expenses, and exception policies. Reusing this example is fine; the assignment does not require inventing a profitable strategy.
Draw arrows between the steps and identify the material required to connect each one. Add comparison methods, records showing what was actually calculated, and conditions to stop or revise. Ask another person to follow the specification and compare orders and cash. A disagreement should first improve the definitions, rather than start a contest over which curve looks better.
Make rules inspectable before asking whether they help
Quantitative research makes judgments repeatable from stated inputs, recording signals, targets, and actual holdings separately. A reconciled account establishes consistency for that calculation. Automation, complexity, and accuracy alone do not establish investment value.
Then select materials and methods for the question, investigating benchmarks, costs, and changing conditions. The small account supplies a checking method; a real strategy still needs its own evidence. Keeping gaps visible helps decide whether the next action is obtaining data, clarifying a rule, or pausing the work.
References
- CFA Institute: Backtesting & Simulation, 2026 public overview. Hypotheses, rules, portfolio formation, and evaluation; no paid course reviewed.
- SEC: Investor Bulletin—Performance Claims, 2022. Costs, methods, and hypothetical performance.
Shared C4 synthetic inputs and checks. Sources checked October 30, 2026. 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 in this phase. No full real C2 backtest or actual trading validation is supplied.