# Research Writing and Decision Communication Present trading research so another person can challenge and reproduce it. Use this workbook alongside the course. Write your answers before opening the solutions. Practical work is self-reviewed; scored knowledge checks are in the Academy. ## 1. Question design ### Economic rationale An economic rationale explains why a relationship might exist, including who could supply or demand the exposure. A plausible story motivates a test; it is not evidence that the result is present or exploitable. ### Falsifiable hypothesis A falsifiable hypothesis specifies a measurable claim and conditions that would contradict it. Avoid wording that can reinterpret every outcome as confirmation. Define the direction, horizon and comparison where relevant. ### Decision relevance Decision relevance connects the study to an actual choice, such as whether to continue researching a rule or reject an execution assumption. A statistically interesting pattern may not change any useful decision. ### Scope and exclusions Scope and exclusions tell the reader where the claim applies and where it does not. Identify instrument, data period, implementation assumptions and omitted cases before presenting the headline conclusion. ### Worked example A memo says the strategy works in the right conditions but never defines those conditions. The claim cannot be falsified because every loss can be labelled the wrong condition afterward. ### Independent exercise Rewrite that claim with an explicit sample, outcome measure and rejection criterion. My inputs and assumptions: My calculation or decision: Evidence that would change my conclusion: ## 2. Evidence presentation ### Dataset provenance Dataset provenance records where observations came from, their timestamps and transformations. A reader should be able to trace a chart point back to source data. State licensing or access constraints that affect reproduction. ### Units and chart labels Every chart needs units, period and an honest scale. Distinguish cumulative cash P&L, percentage return and risk-normalized outcomes. Similar-looking curves can represent different quantities. ### Uncertainty Present uncertainty alongside estimates and explain the method. Avoid phrasing a point estimate as an exact future expectation. The number of decimals should not exceed the practical precision supported by the data. ### Negative findings Negative findings and failed variants are part of the research record. Reporting them helps readers understand selection and prevents others from repeating a rejected approach without learning why it failed. ### Worked example A chart titled performance rises from 100 to 150 but does not say whether it is an index, account balance or cumulative profit. The visual alone cannot establish a 50% strategy return. ### Independent exercise Write a complete caption for a hypothetical net equity curve including cash-flow treatment and observation frequency. My inputs and assumptions: My calculation or decision: Evidence that would change my conclusion: ## 3. Argument quality ### Correlation versus mechanism Correlation and mechanism are different claims. A model can predict without identifying a cause, and a causal explanation may not yield an executable forecast. State which claim the evidence is intended to support. ### Alternative explanations Consider alternative explanations, including common drivers, timing artifacts and data selection. A good memo identifies evidence that could distinguish them instead of presenting the preferred narrative as the only possibility. ### Selection bias Selection bias can arise from chosen dates, surviving accounts, favored instruments or discarded observations. Document the sampling process and assess whether the claim depends on it. ### Unsupported precision Unsupported precision appears in exact targets, probabilities or allocation claims without adequate estimation. Replace certainty with a bounded statement of what the analysis supports and what remains unknown. ### Worked example Two assets move together after a common policy announcement. A memo claiming one caused the other has ignored a plausible common driver unless the design provides additional identification. ### Independent exercise Rewrite that conclusion as a descriptive finding and name the extra question required for a causal claim. My inputs and assumptions: My calculation or decision: Evidence that would change my conclusion: ## 4. Review process ### Reproduction instructions Reproduction instructions should specify data versions, code or formulas, configuration and the sequence of steps. A reviewer should not have to guess a hidden parameter to obtain the reported result. ### Reviewer questions Reviewer questions should focus on material assumptions and discrepancies. Distinguish a request for clarity from a demonstrated error. Resolve disagreements with evidence and an explicit change record. ### Change log A change log states what changed, why and which outputs were affected. Preserve the prior result when useful for understanding the correction. Silent revisions make it difficult to assess whether a conclusion was selected retrospectively. ### Decision and follow-up evidence The final decision should name the action justified by the evidence and the follow-up needed. Continue research, reject the hypothesis and retain a baseline are valid outcomes; the memo need not end with a trade recommendation. ### Worked example A reviewer reproduces a different net result because the original report omitted financing. The correction should update the calculation and conclusion, not merely add a footnote while retaining the old headline. ### Independent exercise Write a correction note containing the error, affected outputs, revised conclusion and a check that prevents recurrence. My inputs and assumptions: My calculation or decision: Evidence that would change my conclusion: ## Course project Submit a concise research memo with evidence, assumptions and a rejection case. ### Self-review rubric - Concepts and reasoning: 25% - Calculations, data and evidence: 30% - Process and risk controls: 25% - Limitations and communication: 20% Record one correction and one next practice task. This rubric is not automatically graded. ## Worked solutions ### Exercise 1 A testable statement names the eligible observations and net metric, then specifies what result would fail the hypothesis. The criterion must precede inspection rather than be chosen to fit the eventual outcome. ### Exercise 2 A useful caption names the account or normalized base, dates, currency, included costs, external-cash-flow treatment and marking frequency. It should identify whether the values are simulated or observed. ### Exercise 3 State the observed co-movement over the specified window and note the shared announcement. A causal claim requires a design or assumptions that separate the assets' influence from the common driver and other confounders. ### Exercise 4 Name the omitted financing, reconcile the revised net result, reassess the decision and add an accounting check covering that cost. Preserve enough history to show why the original conclusion changed. ## Further reading - https://www.itl.nist.gov/div898/handbook/ - https://scikit-learn.org/stable/modules/cross_validation.html