Lin has saved twenty companies and asked AI for thirty strategy ideas. The weekend ends with more material and none of the original questions answered. She may not need another idea. She needs to decide what to investigate in her next two hours.
If every appealing thought immediately deserves a deep investigation, research becomes an expanding to-do list. We will turn two ideas into small projects, starting one and deferring the other. Approving a research project means choosing a use for time and evidence gathering. It does not mean approving an investment of capital.
What you will be able to do
- Narrow “Will this stock rise?” into a question with a subject, mechanism, evidence, and deadline.
- Write company and strategy plans with time limits, cost limits, and stopping conditions.
- Preserve deferred and unsuccessful work, distinguishing a justified revision from rewriting the original test.
A broad idea needs a small first step
An idea might begin with a price change, a business development, or an observed rule. A company announces higher prices, an asset appears less expensive after declining, or recent relative strength seems to persist. Those observations can suggest research without establishing a purchase case. The next step is identifying a question that available material might answer.
“Did the price increase improve profit?” differs from “Is the stock worth buying?” The first requires quantities, prices, costs, and profit measures. The second adds valuation, expectations, and risk. Putting the entire investment decision into the first task makes it harder to choose a starting document or say when that task is finished.
Ask a question worth answering even if you do not trade today. Specify which judgment the answer would change: continuing the investigation, ruling out an object, or choosing the next material. Curiosity is legitimate, but an interesting question that changes none of these decisions need not receive the highest priority in limited research time.
There is no precise universal score for ranking these ideas here. Data availability, personal goals, and the potential value of a judgment cannot necessarily be added together meaningfully. Check essential conditions first, then compare uses of time. You should be able to explain the choice without hiding it behind an unexplained total score.
A company project: what did higher prices leave behind?
People, companies, amounts, and budgets in this chapter are independent synthetic inputs, unrelated to earlier cases or real securities. Qinglan Tools announces that the average price of a product category rose from $100 to $110. Lin's first note says, “The company has pricing power and is worth buying.” That note jumps over several questions.
Her first-stage question becomes: “Within a consistently defined disclosed product category, how did the higher price accompany changes in volume, costs, and gross profit? How far does the available information support a conclusion?” Pricing power remains a proposed mechanism rather than a result the project is committed to proving.
Use teaching figures to see why the other fields matter. The earlier period sold 1,000 units for $100 each, with unit costs of $70. The current period sold 900 for $110, with unit costs of $80. Revenue fell from $100,000 to $99,000. Gross profit fell from $30,000 to $27,000. A 10% selling-price increase did not guarantee higher total gross profit.
Trace the calculation rather than stopping at the price percentage. Earlier gross profit per unit is $100 minus $70, or $30; multiply by 1,000. Current gross profit per unit is $110 minus $80, again $30, but only 900 units sold. The case keeps the inputs simple enough to expose the missing volume and cost questions.
These calculations do not prove that a real company's price increase failed. Product mix, discounts, production, and cost allocation can change comparability. Their purpose is to identify fields Lin needs, without turning a hypothetical result into evidence about an actual business. If the company's published category differs, she must investigate that difference before combining figures.
The first stage has three jobs: locate metric definitions in original reports, verify a comparable scope, and save traceable changes in revenue and costs. EDGAR provides an entry point for U.S. company filings; identify the appropriate issuer and disclosures. Finding a report does not guarantee it contains product-level information. [1]
Lin allocates two hours and no new data expense. Half an hour checks the source, an hour checks the fields, and half an hour organizes supporting and contrary evidence. This is a budget for the exercise, not a standard duration for researching any company. It should make the next action specific without promising a complete investment conclusion in that time.
She writes stopping conditions before starting. Without a consistent scope, she will not calculate a number presented as comparable. If the public documents give only totals, she will mark the product-level question unverified and identify what is missing. Stopping that question does not establish that the company lacks investment value.
The deliverable is a one-page business-verification record. Success means explaining the evidence and gaps, not proving pricing power or reaching a buy decision. Even a verified decline in gross profit leaves questions about its causes and what the market price already assumes. It cannot automatically become a sell recommendation.
A strategy project: before adding parameters, find the inputs
The other idea comes from a historical chart: assets with strong past returns seem to rise again next month. Unlike the company question, this concerns a set of objects under repeatable rules. A chart can motivate the question without supplying the data and specification needed to test it.
The proposed teaching plan fixes a U.S.-listed common-stock universe in advance. After the close on each month's last trading day, it would rank stocks on a previously defined past-return measure, form a candidate portfolio, attempt implementation in the next tradable session, and hold for a month. It would compare the result with a preselected benchmark for the same universe.
This is still not ready to run. The ranking window, exclusions, portfolio weights, missing-price policy, dividend convention, costs, and rebalancing must be specified. Choosing the most attractive version after seeing its chart would change the test. No recommended parameter set or ready-to-trade strategy is provided here; this is a project specification with unresolved fields.
The first stage therefore becomes a data-feasibility check. Can Lin obtain price and corporate-action records available at the relevant time, historical security coverage, and information about delisted companies? Today's surviving-stock list may omit former candidates. It cannot silently represent the full historical set from which a decision could have been made.
She gives this check at most two hours and a $20 budget. A limit is not a requirement to spend it. Reliable free material may answer the question. If only a demonstration sample is available, it can support learning the calculation process, while the report explicitly limits what it can establish about an investment strategy.
Sample use and implementation conditions also need advance planning. Data inspected while exploring and choosing a rule cannot simply be reused as independent proof. Later quantitative chapters examine time divisions and multiple attempts. Merely reserving one interval does not guarantee that every source of selection bias has disappeared.
Research on backtest overfitting explains how repeated selection on finite data can mistake chance patterns for discoveries. Here, apply the basic recordkeeping implication: what versions were tried, why they changed, and what results had already been seen. You are not required to calculate the paper's statistical measure at this stage. [2]
Why only one project starts this week
Lin has three hours this week. The company task requires two, and the strategy feasibility check also requires two. Starting both exceeds the budget. Calling each an easy extra task would not remove the fourth hour; it would merely hide the tradeoff from the plan.
Assume the company reports have already been obtained and their product scope may be checkable. The strategy currently has only a screenshot, without the supplier's data documentation. Lin starts the company task and marks the strategy awaiting material. This choice reflects a clearer immediate deliverable, not a prediction that company research will earn more.
The remaining hour is reserved for review and planning. If the company's scope proves inconsistent, she saves the gap and stops within the two-hour limit. An hour already spent cannot be recovered, and the next hour still competes with other uses. Extending the project through the whole weekend needs a new reason, rather than the mere fact that she began it.
Priorities can change. If usable strategy-data documentation arrives while essential company information remains unavailable, she can reconsider. Record the date, old arrangement, new evidence, and new budget, preserving the first version. A research plan should respond to information while letting another person trace why it changed.
A useful stopping condition concerns the project
“Stop researching if it loses money” confuses a research question with live trading. A project can stop before any trade because data, mechanism, or implementation assumptions fail. A short-term trading loss, on the other hand, does not necessarily reject every underlying hypothesis. Define the condition for the work you are actually doing.
Useful triggers include unavailable required inputs, incompatible definitions, infeasible assumed trade size, reasonable cost scenarios that remove the question's value, or an exhausted budget without an answer. When a trigger occurs, save the record and choose among narrowing, waiting for material, or ending this round. The stopping condition should lead to a named next status.
Guard against changing the question after an unfavorable result. “Predict next month's return” cannot quietly become “explain the long-term trend” and retain a success label. You can open the second question as a new version. The original question still needs its actual outcome, including unresolved or negative findings.
Keep started, awaiting-material, ended, and not-currently-studied ideas on the candidate list. Saving only successful topics makes past judgment appear more insightful than it was. Deferred records can reveal recurring information gaps and help decide which learning or data purchase would be useful, rather than repeatedly restarting the same unanswerable question.
Make Lin's next decision yourself
Explain why the company deliverable is a check of operating changes rather than proof of pricing power. Then show how a three-hour budget should record two tasks requiring two hours each. Write the choice instead of pretending the combined plan takes three hours.
The first answer should allow evidence to weaken or reject the hypothesis. Requiring proof encourages selective support. For the budget, start one project or explicitly narrow a stage, recording the deferred work's reason and restart conditions. Do not conceal the missing fourth hour. A smaller scope must state a smaller deliverable too, rather than keep the original promise with less recorded time.
Change the information gap
You have spent one hour on the company and discover that product volumes are not disclosed. Three more hours could collect online estimates, but their definitions and sources are unclear. This round's limit is still two hours. Would you continue, narrow, or stop? Name a concrete deliverable and a conclusion you cannot make.
A reasonable answer uses the remaining time to record aggregate changes supported by public material, then ends product-level verification with the missing volumes and conditions for obtaining them. Online estimates can remain leads, without becoming verified inputs. They cannot establish that the price increase certainly succeeded or failed. New evidence can justify a later version while preserving this round's unresolved result.
Save one start and one deferral
Choose two ideas from your list. For each, record the subject, question, possible mechanism, competing explanation, original-material entry point, required fields, comparison, time and cost limits, deliverable, and stopping conditions. The company task needs its business scope; the strategy task needs its rules and data prerequisites.
Choose one to start this round and give the other a reason for deferral, missing conditions, and a review trigger. These are the core fields of a one-page project record. Hide the phrase “might make money.” Can someone still identify what you will open and check in the next hour, and where you will stop? If not, narrow the plan again.
Turn interest into the next piece of evidence
Ask which judgment an idea could change, then reduce it to obtainable material and a checkable deliverable. Write mechanisms, alternatives, and inputs together so unfavorable evidence has a place. Budgets and stopping conditions keep the investigation bounded.
Preserve question and experiment versions as work proceeds. New information can change priorities, while observed results cannot be repackaged as the original plan. A useful project can end without further investment when its record explains what was answered, what remains unknown, and what would justify reopening it.
References
- Investor.gov: Using EDGAR to Research Investments. Filing access does not guarantee product-level fields are available.
- Bailey and colleagues: The Probability of Backtest Overfitting, February 2015 revision. The introduction supports repeated-testing and selection concerns; this chapter does not implement its statistical method. Sources checked October 30, 2026.
The plans, company, rules, and budgets are original teaching designs, not a standardized process mandated by the SEC or the paper. AI assisted with writing and checks. A separate AI reviewer has completed an independent full-text review. Human financial review and target-reader testing remain pending.