How to build a data analysis course: a complete guide for lecturers

Central question: what decision is being made, and what evidence would actually answer it?

A strong Data Analysis course begins before the dataset opens. Students need to translate a broad managerial question into a unit of analysis, outcome, explanatory variables, population, time horizon and decision threshold. This prevents the common failure of producing technically polished analysis that answers the wrong question.

What this concept covers

• Decision questions versus analytical questions.
• Units of analysis, target populations and time horizons.
• Outcome, predictor and control variables.
• Descriptive, inferential, predictive and causal objectives.
• Analysis plans, success criteria and stopping rules.

Learning outcomes

By the end of this concept, students should be able to:

• Translate an ambiguous business problem into an answerable analytical question.
• State the unit of analysis and target population.
• Distinguish descriptive, explanatory, predictive and causal goals.
• Specify evidence that could change the decision.

How to teach it

• Give three vague executive questions and ask teams to rewrite them as testable analysis briefs.
• Require a one-page analysis plan before students see the data.
• Ask students to state what result would change their recommendation and what result would not.

Case-style example

Northstar Grocers is considering a new loyalty feature. Store A reports a 6% sales increase after launch, but the feature was introduced only to already-active members. Students must write the analysis brief: define the customer-level outcome, comparison group, observation window and decision threshold. One team may prioritise incremental revenue; another may prioritise repeat visits. They discover that a precise method cannot rescue an ambiguous decision question.

Where students struggle

Students often start with columns rather than the decision. They need to see that method choice depends on what kind of claim the decision requires.

Reading and a quick check

Use the opening chapters of the core textbook on data, variables and statistical studies. Quick check: ask students to label five claims as descriptive, inferential, predictive or causal and explain what evidence each needs.

Where it leads

Once the question is explicit, students can judge whether the available data were measured well enough to answer it.

Central question: are these variables measured consistently enough for the analysis we want to perform?

Data quality is not a cleaning footnote. Measurement definitions, coding choices, missing values, units, duplicates and inconsistent time periods can change the substantive conclusion. Lecturers should teach a data dictionary and a quality audit as analytical artefacts, not administrative chores.

What this concept covers

• Categorical, ordinal, interval and ratio variables.
• Operational definitions and measurement error.
• Missingness, duplicates, impossible values and inconsistent units.
• Data dictionaries, provenance and lineage.
• Validity, reliability and fitness for purpose.

Learning outcomes

By the end of this concept, students should be able to:

• Classify variables by type and measurement scale.
• Diagnose data-quality problems that threaten a decision.
• Write an operational definition that another analyst could reproduce.
• Build a concise data-quality report and prioritise fixes.

How to teach it

• Provide a deliberately messy data dictionary and ask students which fields can support the proposed metric.
• Make students separate data-quality issues that can be repaired from design flaws that cannot.
• Require a short provenance note with every dataset used in assessment.

Case-style example

Harbor Health has 20,000 appointment records. Waiting time appears to fall from 31 to 24 minutes, but 18% of rows lack arrival timestamps, two clinics record seconds rather than minutes, and repeat appointments share the same encounter ID. Students must quantify each issue, decide which rows can be corrected and calculate the trend under two defensible cleaning rules. They discover that data quality choices can be decision-critical, not cosmetic.

Where students struggle

Students often assume a spreadsheet is an objective record. They need to ask how each field was produced, what it excludes and whether the definition changed over time.

Reading and a quick check

Assign a short data-governance or quality note alongside the textbook section on data types. Quick check: give six suspect values and ask whether each is invalid, missing, unusual or merely inconvenient.

Where it leads

After measurement quality, the next question is whether the observations represent the population the lecturer wants students to discuss.

Central question: who is represented in the dataset, who is missing, and how could the sampling process bias the conclusion?

Sampling connects a dataset to a population. Students should learn that a large sample can still be biased, a convenience sample can still be useful for a bounded claim, and nonresponse can matter more than sample size. Business datasets are often generated by processes rather than probability samples, so the course should make selection mechanisms visible.

What this concept covers

• Populations, samples and sampling frames.
• Random, stratified, cluster and convenience sampling.
• Coverage error, nonresponse and survivorship bias.
• Selection into administrative and digital datasets.
• External validity and transportability of conclusions.

Learning outcomes

By the end of this concept, students should be able to:

• Design a sampling strategy for a stated decision.
• Identify coverage and selection biases in observed data.
• Explain how sample size and representativeness differ.
• Bound claims when the sample cannot represent the full population.

How to teach it

• Compare an online voluntary survey with a stratified customer sample.
• Ask students to draw the selection process that generated an administrative dataset.
• Use weighting as a concept, but make students explain what weighting cannot fix.

Case-style example

MetroBank wants to estimate satisfaction across 400,000 customers. It receives 8,400 app-survey responses, 72% from customers under 40, while that group is only 39% of the base. Students must compare the raw result with a post-stratified estimate and write a two-sentence limitation. One group may weight by age only; another may demand channel and product strata. They discover that more observations do not automatically mean more representative evidence.

Where students struggle

Students equate big data with representative data. The teaching priority is the mechanism that determines who enters the dataset.

Reading and a quick check

Read the textbook chapter on sampling and data collection. Quick check: ask students what population can defensibly be described by a dataset of active app users and what population cannot.

Where it leads

Once students understand provenance and selection, they can prepare the data without hiding analytical choices inside ad hoc spreadsheet edits.

Central question: how can we transform raw data into analysis-ready data without losing the ability to explain what changed?

Cleaning should be teachable as a sequence of explicit decisions: preserve raw data, define rules, transform reproducibly, validate outputs and record exceptions. Whether the tool is Excel, R or Python, the principle is the same. A second analyst should be able to reconstruct the analysis dataset from the raw source and the documented steps.

What this concept covers

• Raw, interim and analysis-ready data.
• Type conversion, recoding, joins, reshaping and derived variables.
• Missing-data handling and outlier rules.
• Validation checks and reconciliation totals.
• Scripts, notebooks, formulas and change logs for reproducibility.

Learning outcomes

By the end of this concept, students should be able to:

• Design a reproducible cleaning workflow.
• Justify missing-data and outlier treatment.
• Validate joins, transformations and derived variables.
• Create a compact audit trail linking raw data to final analysis.

How to teach it

• Give two tables with duplicated keys and ask students to predict what a join will do before executing it.
• Require a validation table with row counts and control totals after each major transformation.
• Mark the clarity of the cleaning log, not only the final clean file.

Case-style example

FieldWorks has 52,300 sales lines and 8,700 customer records. A many-to-many join unexpectedly creates 64,900 rows because 312 customer IDs are duplicated. Students must diagnose the join, choose a defensible de-duplication rule, reconcile revenue to the raw ledger and document the decision. One approach keeps the most recent customer record; another aggregates stable attributes. They discover that silent transformation errors can create more analytical damage than an incorrect formula.

Where students struggle

Students often overwrite raw data, delete inconvenient rows and forget why. Reproducibility turns those actions into reviewable analytical decisions.

Reading and a quick check

Use a short reproducibility checklist and a textbook section on data preparation. Quick check: ask students to state the three controls they would run after a merge.

Where it leads

With analysis-ready data, students can begin exploring distributions, variability and patterns before formal modelling.

Central question: what does the dataset look like before we make formal claims about it?

Exploratory data analysis teaches students to look before they model. Location, spread, distribution shape, skewness, outliers, groups and time patterns should be used to generate questions and identify risks. The distinction between describing what is observed and inferring beyond the sample should remain explicit.

What this concept covers

• Means, medians, quantiles, variance and standard deviation.
• Distribution shape, skewness and outliers.
• Grouped summaries, cross-tabulations and segment comparisons.
• Time trends, indices and growth rates.
• Exploratory analysis as a generator of hypotheses rather than proof.

Learning outcomes

By the end of this concept, students should be able to:

• Select descriptive statistics that fit a variable and decision.
• Compare groups without hiding distributional differences.
• Identify unusual observations and trends worth investigating.
• Write an exploratory finding without overstating inference.

How to teach it

• Give the same dataset to different groups and require one numerical and one visual summary.
• Ask students to explain why median can outperform mean for a skewed variable.
• Use the Financial Statement Analysis simulation after students can interpret trends and ratios.

Case-style example

Fjord Retail reports average order value of £86 across 12,000 orders, but the median is £43 and the top 2% of orders contribute 24% of revenue. Students must compute quartiles, trimmed means and segment summaries, then decide whether a premium-customer campaign is justified. One group may focus on concentration; another may focus on broad customer value. They discover that a single average can conceal the commercial structure of the data.

Where students struggle

Students can calculate summary statistics without asking which one answers the decision. Push them to connect each measure to shape, scale and business consequence.

Reading and a quick check

Read the descriptive-statistics chapter in the core textbook. Quick check: ask which statistic should be reported for a heavily right-skewed customer-spend distribution and what information would be lost.

Where a simulation fits

The Financial Statement Analysis Simulation fits after this concept because students already know how to read changing ratios and trends. It gives them repeated reporting periods, updated information and an individual evidence-based judgement rather than a static worksheet.

Where it leads

Exploration becomes more useful when students can communicate patterns clearly and avoid visual designs that create false certainty.

Central question: which visual representation makes the comparison or decision easiest to see without distorting the evidence?

Visualisation should be taught as analytical communication, not decoration. Students need to choose encodings that match the task, preserve scale, show uncertainty where relevant and foreground the comparison the audience must make. A chart should make reasoning easier while leaving enough context to audit the claim.

What this concept covers

• Position, length, area and colour as visual encodings.
• Bars, lines, distributions, scatterplots and small multiples.
• Scales, baselines, binning and annotation.
• Uncertainty, sample size and contextual reference points.
• Dashboard and executive-summary principles.

Learning outcomes

By the end of this concept, students should be able to:

• Choose a chart form that matches the comparison.
• Detect misleading axes, aggregation and visual clutter.
• Annotate a visual to make the analytical claim explicit.
• Design an executive view that preserves relevant uncertainty.

How to teach it

• Give students a misleading chart and require a redesign plus a written explanation of what changed.
• Have groups create two charts for the same data: one exploratory and one decision-facing.
• Mark titles as claims that must be supported by the visual.

Case-style example

Aster Mobility shows monthly complaints falling from 110 to 92 while customers grow from 5,000 to 8,400. Students must build one chart using counts and another using complaints per 1,000 customers, then recommend which belongs in an executive report. A raw-count view suggests modest improvement; a rate view suggests a much larger change. They discover that denominator choice can be more important than chart polish.

Where students struggle

Students over-focus on aesthetics or default chart types. The key question is whether the visual makes the intended comparison honest and fast.

Reading and a quick check

Assign Oral et al. on visualisation for decision-making. Quick check: ask students to identify the decision, comparison, denominator and uncertainty cue in a published chart.

Where a simulation fits

The Financial Statement Analysis Simulation can be debriefed here by asking students to redesign one ratio or trend view that would make their final judgement easier to defend.

Where it leads

Visual patterns lead naturally to probability, because students now need a language for variability and uncertainty rather than only observed differences.

Central question: how should we reason about variation and risk when outcomes are not certain?

Probability provides the language that connects observed variability to inference and decisions under uncertainty. The goal is not to turn the course into a probability-theory module. Students should be able to model events, conditional information and simple random processes well enough to interpret risk, expected value and sampling variation.

What this concept covers

• Events, complements, unions and intersections.
• Conditional probability, independence and Bayes logic.
• Random variables, expectation and variance.
• Common distributions as models rather than labels.
• Monte Carlo simulation as a way to propagate uncertainty.

Learning outcomes

By the end of this concept, students should be able to:

• Calculate and interpret simple conditional probabilities.
• Explain dependence and base-rate effects.
• Use expectation and variability to compare uncertain outcomes.
• Simulate a simple decision to reveal a distribution of possible results.

How to teach it

• Use a medical-test or credit-screening example to expose base-rate neglect.
• Build a simple spreadsheet Monte Carlo model with uncertain demand and margin.
• Ask students to report a range and probability, not only an expected value.

Case-style example

Verdant Energy is considering a pilot with annual contribution of either £1.8m, £0.8m or -£0.6m with probabilities 0.25, 0.50 and 0.25. A weather shock is correlated with the downside state. Students must compute expected contribution, model 5,000 simulated outcomes for a two-year rollout and compare a risk-neutral recommendation with a 10% downside-probability limit. They discover that the same expected value can support different decisions once risk tolerance is explicit.

Where students struggle

Students confuse probability with certainty and often treat modelled probabilities as observed facts. Keep assumptions visible and ask what evidence would justify them.

Reading and a quick check

Use the core textbook probability chapters. Quick check: ask why multiplying two marginal probabilities can fail when events are dependent.

Where a simulation fits

The Portfolio Management Simulation can later extend this logic into risk-return trade-offs, but only after students have the simulation-specific prerequisites in CAPM, Sharpe Ratio and mean-variance optimisation.

Where it leads

Probability prepares students to explain how sample estimates vary and why an estimate should be accompanied by a measure of uncertainty.

Central question: what can a sample tell us about a population quantity, and how uncertain is the estimate?

Estimation is the bridge from sample description to population reasoning. Students should understand standard errors, confidence intervals and the role of sample size as statements about procedures and uncertainty, not magical ranges that guarantee a parameter lies inside a particular interval.

What this concept covers

• Parameters, estimators and sampling distributions.
• Standard error and the effect of sample size.
• Confidence intervals for means, proportions and differences.
• Margin of error and precision.
• Practical interpretation of uncertainty intervals.

Learning outcomes

By the end of this concept, students should be able to:

• Construct and interpret confidence intervals.
• Explain how sample size affects precision.
• Distinguish parameter uncertainty from individual outcome variability.
• Use interval width and decision thresholds together.

How to teach it

• Simulate repeated samples so students see the long-run coverage idea.
• Give two estimates with different sample sizes but similar point estimates and ask which is more decision-useful.
• Require an interval next to every key estimate in a class exercise.

Case-style example

SeaGlass Hotels surveys 420 guests and estimates that 61% would use mobile check-in, with a 95% confidence interval of roughly 56%-66%. The rollout threshold is 60%. Students must decide whether to launch, pilot or collect more data, then calculate the sample size needed to reduce margin of error. One group may accept the expected upside; another may demand greater precision. They discover that uncertainty can change the action even when the point estimate crosses a target.

Where students struggle

Students often describe a confidence interval as the probability that the fixed parameter lies in that realised interval. Focus on the procedure and decision implications rather than a memorised sentence.

Reading and a quick check

Use the core textbook chapter on interval estimation. Quick check: ask what happens to interval width when sample size quadruples, all else equal.

Where it leads

Once students can estimate with uncertainty, they are ready to evaluate explicit claims and experimental comparisons.

Central question: what evidence would count against a claim, and does the result matter in practice as well as statistically?

Hypothesis testing should be taught as one component of evidence, not a binary truth machine. Students need to connect the null model, test statistic, p-value, effect size, confidence interval and practical threshold. A/B testing adds design discipline: randomisation, pre-specified metrics, power, interference and multiple testing.

What this concept covers

• Null and alternative hypotheses.
• Test statistics, p-values and Type I/II errors.
• Effect sizes, confidence intervals and practical significance.
• Randomised experiments and A/B tests.
• Power, multiple testing, peeking and pre-specification.

Learning outcomes

By the end of this concept, students should be able to:

• Choose a test that matches the design and variable type.
• Interpret p-values without converting them into probabilities that a hypothesis is true.
• Compare statistical and practical significance.
• Design an A/B test with a primary metric and decision rule.

How to teach it

• Use a small experiment where the p-value and business threshold point in different directions.
• Ask students to pre-register the metric, unit, horizon and stopping rule before revealing results.
• Debate whether a statistically detectable effect is worth implementation cost.

Case-style example

CloudCart tests a new checkout on 40,000 sessions. Conversion rises from 4.80% to 5.02%; the estimated lift is positive but small, and implementation costs £180,000. Students must compute the absolute and relative lift, inspect the confidence interval and compare annualised contribution against cost. One group may launch on strategic grounds; another may reject the economics. They discover that statistical significance and decision significance answer different questions.

Where students struggle

Students chase p < 0.05 and ignore effect size, design validity and multiplicity. The better habit is to ask what decision rule was set before the result appeared.

Reading and a quick check

Assign Wasserstein, Schirm and Lazar (2019) plus a selected section from Kohavi, Tang and Xu (2020). Quick check: ask students to rewrite a sentence that incorrectly says a p-value is the probability the null hypothesis is true.

Where it leads

Testing compares claims; regression then lets students estimate relationships while controlling for additional variables and making the model assumptions explicit.

Central question: how strongly are variables related, what does the model estimate, and what claims are not justified by the design?

Regression is often the technical centre of a Data Analysis course because it connects description, inference and prediction. The teaching challenge is to keep coefficients substantive: units, reference groups, uncertainty, functional form, diagnostics and omitted variables matter more than simply producing a regression table.

What this concept covers

• Covariance and correlation.
• Simple and multiple linear regression.
• Coefficient interpretation, indicators and interactions.
• Residuals, non-linearity, heteroskedasticity and influential observations.
• Association versus causal interpretation.

Learning outcomes

By the end of this concept, students should be able to:

• Interpret regression coefficients in units that matter.
• Diagnose important model failures.
• Compare models without treating fit as the only criterion.
• State clearly when a coefficient supports association but not causation.

How to teach it

• Use a dataset where an apparent bivariate relationship changes after adding a confounder.
• Require students to plot residuals and inspect influential points before reporting coefficients.
• Have students write one allowed claim and one prohibited causal claim from the same model.

Case-style example

BrightHire models first-year sales performance for 2,400 recruits using interview score, prior experience and training hours. Interview score has a positive coefficient until region and role type are added, after which it shrinks by 55%. Students must estimate both models, interpret the change and decide whether interview score should influence hiring. One group may value predictive usefulness; another may reject a causal story. They discover that coefficient stability depends on model specification and design.

Where students struggle

Students read sign and significance but miss units, confounding, non-linearity and leverage points. Ask them what must be true for the coefficient to answer the business question.

Reading and a quick check

Use the regression chapters from the core textbook and ISLP for model diagnostics. Quick check: ask students to interpret a 0.12 coefficient when the predictor is measured in £1,000 units.

Where it leads

Regression becomes a baseline for prediction, where the standard of evidence shifts from explanation to performance on genuinely unseen data.

Central question: will the model perform on new data, and is the validation design credible?

Prediction changes the evaluation question. A complex model can fit the training data extremely well and still fail out of sample. Students should learn train-validation-test separation, cross-validation, baseline comparison, leakage checks, metric choice, calibration and the difference between forecasting a target and explaining its cause.

What this concept covers

• Training, validation and test sets.
• Cross-validation and resampling.
• Overfitting, regularisation and bias-variance trade-offs.
• Classification and regression metrics.
• Forecasting, temporal validation, drift and data leakage.

Learning outcomes

By the end of this concept, students should be able to:

• Design an out-of-sample evaluation that matches deployment.
• Compare a complex model with a simple baseline.
• Select evaluation metrics based on the decision cost.
• Detect leakage and explain why it inflates performance.

How to teach it

• Create a leakage trap where a post-outcome variable makes accuracy look exceptional.
• Use time-split validation for forecasting so future observations never enter training.
• Require a model card summarising data, metrics, limitations and failure modes.

Case-style example

ParcelPilot predicts late deliveries using 180,000 historical shipments. A random split gives 94% accuracy, but a time-based holdout drops to 78% because a route-status field is only finalised after delivery. Students must remove the leaked field, compare logistic regression with a tree model and choose a metric where false negatives cost four times false positives. They discover that validation design can matter more than algorithm sophistication.

Where students struggle

Students optimise a score without matching the validation scheme to the real future decision. Make them draw the deployment timeline and identify what information will actually exist at prediction time.

Reading and a quick check

Assign Kapoor and Narayanan (2023), Rudin et al. (2022) and selected ISLP chapters. Quick check: ask why a shuffled random split can be invalid for a time-series forecast.

Where a simulation fits

The Portfolio Management Simulation works as a specialist applied quantitative lab after students have basic CAPM, Sharpe Ratio and mean-variance optimisation. Teams use historical data and a model, then rebalance as new information arrives. The value for a Data Analysis course is the gap between model output and defensible decision, not general software practice.

Where it leads

The final concept asks students to make analytical work reviewable, fair, reproducible and useful in organisational decisions.

Central question: what makes an analytical recommendation trustworthy enough to act on and accountable enough to review?

The course should end by integrating technical quality with governance. Students need to document provenance, assumptions, model limits, privacy or fairness risks, AI assistance and reproducibility, then communicate a recommendation that a decision-maker can challenge. A strong final output distinguishes evidence, judgement and policy choice.

What this concept covers

• Reproducibility, versioning and analytical audit trails.
• Privacy, fairness, representativeness and stakeholder impact.
• Model documentation and monitoring.
• Permitted and declared use of generative AI.
• Decision memos that separate evidence, assumptions, uncertainty and recommendation.

Learning outcomes

By the end of this concept, students should be able to:

• Produce a reproducible analysis package that another analyst can review.
• Identify ethical and governance risks in data or model use.
• Declare AI assistance and verify generated code or claims.
• Defend a recommendation while naming uncertainty and missing information.

How to teach it

• Run a red-team review where another group tries to reproduce and challenge the analysis.
• Require an AI-use declaration with the final project.
• Use a one-page decision memo that separates facts, assumptions, analysis and recommendation.

Case-style example

NovaLend builds a default-risk model with 72,000 applications. Overall accuracy is 86%, but approval rates differ sharply by postcode group and 14% of historical outcomes are missing after customers refinance elsewhere. Students must audit the missingness, compare subgroup error rates, propose monitoring metrics and write a launch recommendation. One group may favour a controlled pilot; another may pause. They discover that technically strong performance can still be insufficient for a trustworthy decision.

Where students struggle

Students can treat ethics as a final discussion disconnected from method. Integrate it with provenance, sampling, metric choice, deployment and monitoring so it changes analytical decisions.

Reading and a quick check

Assign the NIST AI Risk Management Framework 1.0 and a reproducibility reading. Quick check: ask students which parts of their capstone another analyst would need to reproduce the result and challenge the recommendation.

Where a simulation fits

The Managerial Accounting Simulation and Capital Budgeting Simulation can be used as short closing decision labs where students combine quantitative evidence with trade-offs and written rationale. They are applied contexts, not replacements for general Data Analysis software.

Where it leads

This completes the lifecycle: students should be able to frame, prepare, analyse, model, communicate and defend an evidence-based decision.

10-week course structure

A 10-session version should compress adjacent stages rather than remove the analytical lifecycle. Keep data provenance, inference, regression, validation and governance visible.

14-week course structure

A 14-session version creates room to separate experiments from hypothesis testing, add more validation work and use a longer applied simulation or capstone review without crowding the core methods.

The Financial Statement Analysis Simulation is the strongest direct fit for a general Data Analysis course because it makes interpretation across changing quantitative information visible. Students work individually as Financial Analysts, connect the Income Statement, Balance Sheet and Cash Flow Statement, calculate and interpret profitability, liquidity, efficiency and solvency ratios, then update their judgement across five reporting periods.

Why it fits here

By the end of Session 5, students should know how to summarise distributions, compare periods, interpret ratios and distinguish a numerical change from a meaningful business change. The simulation then turns a static analysis exercise into repeated evidence updating.

What students do

• Begin with an introductory video and mission as an individual Financial Analyst.
• Analyse the three primary financial statements and calculate selected ratios.
• Review updated statements, company developments and additional reports across five evolving reporting periods.
• Identify important trends and warning signs, decide whether prior interpretation still holds and submit a revised share-price prediction.
• Use the final reflection view to compare predictions, submitted answers and correct answers, then explain which calculations or interpretations changed the decision.
• Treat the share-price prediction as an engagement and reflection device based on supplied information, not a full equity-valuation model.

Recommended course placement

• After Session 5 as a 2-3 hour workshop on exploratory financial data and trend interpretation.
• Across Sessions 5 and 6 when the second class is used for visualisation and decision communication.
• As an individual applied evidence point before the main group analytics project.

Knowledge students should have before the simulation

• Basic financial-statement vocabulary is useful, but the product is beginner accessible and provides guidance.
• Students should already be able to calculate or interpret ratios, compare periods and explain how a metric links to a decision.
• No programming knowledge is required; the exercise should be framed as quantitative interpretation rather than software training.

Delivery format

Suggested debrief questions

• Which ratio or trend changed your view most, and why?
• When did a new period justify revising the prior conclusion rather than merely adding noise?
• Did you follow the cohort median too closely at any point?
• Which metric was accurate numerically but easy to misinterpret commercially?
• What additional data would you want before making an investment recommendation?

The Portfolio Management Simulation is a strong quantitative extension for courses that want a finance-analytics pathway. Teams manage equity portfolios across 25 global companies using CAPM and mean-variance optimisation, then make buy, sell and rebalance decisions as market information changes. Hedge Fund teams seek stronger Alpha; Pension Fund teams seek higher Sharpe Ratio.

Why it fits here

It fits after model evaluation because it makes a useful distinction between model output and action. The model provides expected return, beta, covariance, volatility and portfolio weights; teams still have to decide how much evidence to trust, how concentrated to be and whether new information justifies a rebalance. It should not be run before the specific portfolio prerequisites are in place.

What students do

• Work in teams of 3-5 and receive either a Hedge Fund or Pension Fund mandate.
• Review historical market data for 25 global companies and use the supplied Excel model to estimate expected returns, beta, covariance, volatility and portfolio weights.
• Construct an initial equity portfolio, choosing securities, allocations and cash, with leverage available where the mandate permits it.
• Move through later trading quarters as new company and market information arrives, deciding whether to hold, buy, sell or rebalance.
• Review return, volatility and role-specific performance. Compare Hedge Fund teams with Hedge Fund teams and Pension Fund teams with Pension Fund teams.

Recommended course placement

• After Session 11 when the lecturer wants a specialist finance application of validation, risk, updating and model-vs-judgement.
• Across two classes if the 3-6 hour experience is split, with a short reflection between sessions.
• As an optional pathway rather than a compulsory general Data Analysis activity if the cohort lacks portfolio-theory prerequisites.

Knowledge students should have before the simulation

• Basic CAPM.
• Sharpe Ratio and the meaning of risk-adjusted performance.
• Mean-variance optimisation, expected return, covariance, volatility and portfolio weights.
• Ability to distinguish a model assumption from observed market information.

Delivery format

Suggested debrief questions

• Which modelling assumption had the greatest effect on your initial portfolio?
• What new information was material enough to justify a rebalance?
• Did a high return come from a defensible process or excessive concentration and risk?
• Where did the optimisation model help, and where did judgement override it?
• How would your evaluation change if the same portfolio were assessed under the other mandate?

Grading with the Financial Statement Analysis Simulation

Students work individually as Financial Analysts, calculate and interpret selected ratios, review changing financial information and revise a share-price judgement across reporting periods.

• Accuracy and interpretation of selected ratio calculations.
• Ability to identify material changes across periods rather than list every movement.
• Quality of the evidence used to revise or retain a prediction.
• Reflection on calculation or interpretation errors and what additional data would be needed.

Grading with the Portfolio Management Simulation

Teams use historical data, CAPM and mean-variance optimisation to construct and rebalance portfolios under Hedge Fund or Pension Fund mandates.

• Quality of portfolio construction relative to the assigned mandate.
• Use and interpretation of expected return, beta, covariance, volatility and portfolio weights.
• Reasoning behind major buy, sell, cash, concentration or rebalance decisions.
• Risk-adjusted interpretation of outcomes rather than rewarding final return alone.
• Add an individual viva, memo or reflection because the platform does not attribute team arguments to specific students.