How to build a business statistics course: a complete guide for lecturers

Central question: What must be defined before a dataset can answer a managerial question?

Business statistics should begin with the decision rather than the formula. Students need to identify the population, unit of analysis, variables, measurement scales and data-generating process before they calculate anything. This prevents a technically correct analysis from answering the wrong business question.

What this concept covers

• Managerial questions, statistical questions and decision criteria.
• Populations, samples, observational units and variables.
• Categorical versus quantitative data and measurement scales.
• Primary versus secondary data, administrative data and digital trace data.
• Data quality, missingness, selection bias and the distinction between observed data and the construct a manager actually cares about.

Learning outcomes

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

• Frame a business problem as a statistical question.
• Classify variables and choose suitable summaries.
• Identify the population, sample and unit of analysis.
• Diagnose obvious threats to data quality before analysis.

Supports Course Outcomes 1 and 2.

How to teach it

• Start with a short managerial brief and ask students what they would need to observe before opening a spreadsheet.
• Give deliberately messy variable definitions and ask groups to rewrite them operationally.
• Use a data dictionary exercise so students see how coding choices shape later analysis.
• End with a one-paragraph analysis plan rather than a calculation.

Case-style example

HarbourCart wants to know whether a new delivery promise improves customer retention. It has 18,400 orders, 7,250 customer records and a 90-day repeat-purchase flag, but 14% of customers used guest checkout. Students define the unit of analysis, population, outcome and treatment exposure, then decide whether guest users should be excluded or retained. They discover that the data structure, not the test, determines whether the retention question is answerable.

Where students struggle

Students often treat every row as an independent observation and assume that variables mean exactly what their labels imply. Push them to ask who is missing, how values were recorded and whether the same customer appears more than once.

Reading and a quick check

Use the opening chapters of the core textbook on data, decisions and data sources. Quick check: ask students to write one sentence each for the population, sample, unit of analysis and outcome variable in a fresh business brief.

Applied teaching activity

Run a data-audit workshop in which students receive a one-page dataset description with duplicated IDs, inconsistent categories and missing values. Require a short note separating fixable cleaning issues from limitations that change what can be inferred.

Where it leads

Once the question and data structure are defensible, students can learn how to summarise what the sample actually contains.

Central question: How can a lecturer help students describe a dataset without hiding the pattern that matters?

Descriptive statistics are not a preliminary chore. They are the first model of the business situation. Students should learn to combine tables, graphs and numerical summaries, and to explain why the mean, median, standard deviation or a particular chart is informative for the decision at hand.

What this concept covers

• Frequency tables and proportions for categorical data.
• Mean, median, quantiles, variance, standard deviation and coefficient of variation.
• Histograms, box plots, bar charts, scatterplots and small-multiple displays.
• Skewness, outliers and transformations.
• Segment comparisons and the danger of averaging across heterogeneous groups.

Learning outcomes

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

• Choose appropriate graphical and numerical summaries.
• Explain the business meaning of spread and skewness.
• Identify outliers that require investigation rather than automatic deletion.
• Communicate a descriptive result without overstating what it proves.

Supports Course Outcomes 2 and 3.

How to teach it

• Give two charts of the same data with different scales and ask which better supports the decision.
• Use a small dataset where mean and median imply different stories.
• Require every chart to have a decision-relevant title rather than a variable name.
• Ask students to write a 60-word executive interpretation beneath each visual.

Case-style example

Redwood Hotels reviews 2,400 nightly room rates. The mean is £188, the median is £154 and the 90th percentile is £329 because a small set of suites and event nights are expensive. Students calculate centre and spread, produce a box plot by room type and compare conclusions using the mean versus the median. They discover that a single portfolio-wide average can misrepresent the typical guest price and mask profitable segments.

Where students struggle

Students can become chart decorators, adding visual complexity without improving inference. Mark the match between question, variable type and display, and ask what a manager can conclude that they could not see in the raw table.

Reading and a quick check

Core textbook chapters on categorical and quantitative summaries. Quick check: provide a skewed revenue distribution and ask which single measure of centre they would report to a board and what second statistic they would add.

Where a simulation fits

The Financial Statement Analysis Simulation can fit after descriptive statistics if the lecturer wants an applied interpretation exercise. Students work across multiple reporting periods, calculate ratios and judge whether trends are positive, negative or inconclusive. Use it to discuss change over time and evidence selection, not as a substitute for teaching finance.

Where it leads

The next step is to move from what happened in a sample to formal reasoning about uncertainty.

Central question: How should students reason about uncertain events before they meet statistical inference?

Probability gives students a language for uncertainty. The goal is not to create a combinatorics course, but to make students comfortable with conditional probability, independence, complements and Bayes-style updating in situations such as credit risk, quality, marketing response and fraud detection.

What this concept covers

• Events, complements, unions and intersections.
• Conditional probability and independence.
• Tree diagrams and contingency tables.
• Bayes rule as evidence updating.
• Base-rate effects and false-positive reasoning.

Learning outcomes

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

• Compute and interpret conditional probabilities.
• Distinguish P(A|B) from P(B|A).
• Recognise when independence assumptions are implausible.
• Explain how base rates affect screening and classification decisions.

Supports Course Outcomes 3 and 4.

How to teach it

• Use a fraud-screening example where the test is accurate but fraud is rare.
• Ask students to solve the same problem with a tree and a contingency table.
• Have groups state the operational cost of false positives and false negatives.
• Connect conditional probability to later classification metrics without turning the session into machine learning.

Case-style example

Northstar Bank screens 100,000 card transactions. Fraud occurs in 0.8% of transactions. A rule flags 92% of fraud but also flags 3.5% of legitimate transactions. Students compute the probability that a flagged transaction is actually fraudulent and compare a “high sensitivity” policy with a stricter policy that misses more fraud. They discover why an impressive detection rate can still create a large false-positive workload when the base rate is low.

Where students struggle

The most common error is reversing the condition. Students also treat independence as a mathematical convenience rather than an empirical claim. Require every probability statement to be written in words before calculation.

Reading and a quick check

Core textbook probability chapter. Quick check: ask students to explain, without formulas, why a positive test result can have a low positive predictive value when the event is rare.

Applied teaching activity

Use a short risk-screening decision workshop in which each team chooses a threshold under different false-positive and false-negative costs, then defends the trade-off.

Where it leads

Probability becomes more useful when students model uncertain numerical outcomes with random variables and distributions.

Central question: Which probability model is useful for a business quantity, and what assumptions make that model credible?

Students need to see distributions as models of variation rather than a list of formulas. The teaching emphasis should be on choosing a plausible model, interpreting expected value and variability, and recognising when the normal model is useful or misleading.

What this concept covers

• Discrete and continuous random variables.
• Expected value, variance and standard deviation.
• Binomial, Poisson and normal models in business settings.
• Standardisation and z-scores.
• Distributional assumptions, tail risk and simulation as a way to inspect uncertainty.

Learning outcomes

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

• Match common distributions to suitable business processes.
• Calculate expected values and probabilities.
• Interpret standardised values and tail probabilities.
• Challenge an inappropriate distributional assumption.

Supports Course Outcomes 4 and 5.

How to teach it

• Use demand counts, conversion events and service times to contrast discrete and continuous models.
• Ask students to state why a binomial or Poisson model might fail.
• Use spreadsheet simulation to show repeated outcomes before introducing formal formulas.
• Compare two distributions with the same mean but different variance and ask which a manager should prefer.

Case-style example

MetroBrew sells 420 takeaway coffees per weekday on average, with daily standard deviation 38. A manager proposes staffing to the mean. Students estimate the probability that demand exceeds 480 under a normal approximation, then compare it with an empirical distribution containing event-day spikes. They must recommend staffing using either model and explain the risk of each choice. They discover that a convenient distribution can understate tail demand when the process has special-event regimes.

Where students struggle

Students often assume “looks bell-shaped” is enough evidence for a normal model. Make them connect the distribution to how the data are generated and what tail behaviour matters for the decision.

Reading and a quick check

Core textbook chapters on random variables and continuous distributions. Quick check: give three business processes and ask students to choose a model plus one assumption they would verify.

Where a simulation fits

The Portfolio Management Simulation becomes useful after students understand expected return, variance and co-movement, but it has additional finance prerequisites. If used here or later, provide a short bridge on CAPM, Sharpe Ratio and mean-variance optimisation before launch.

Where it leads

With a model of individual outcomes in place, students can study how statistics behave across repeated samples.

Central question: When does a sample statistic give a stable basis for inference about a wider population?

This concept is the bridge from description to inference. Students should understand random sampling, sampling error and the central limit theorem conceptually, including why larger samples reduce standard error but do not automatically remove bias.

What this concept covers

• Sampling frames, probability samples and convenience samples.
• Sampling error versus non-sampling error.
• Sampling distributions of means and proportions.
• Standard error and the central limit theorem.
• Bias, non-response and the limits of large datasets.

Learning outcomes

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

• Explain what varies across repeated samples.
• Compute standard errors for common statistics.
• Use the central limit theorem appropriately.
• Separate precision from representativeness.

Supports Course Outcomes 5 and 6.

How to teach it

• Simulate repeated samples from a skewed population and plot the sample means.
• Compare a small random sample with a huge biased convenience sample.
• Ask students which problem is fixed by increasing n and which is not.
• Use polling or customer-survey examples where the sampling frame is visibly incomplete.

Case-style example

GreenLoop Energy wants to estimate average monthly household usage from 48,000 smart meters. A random sample of 400 customers gives a mean of 312 kWh, while a voluntary app survey of 6,200 users gives 268 kWh. Students calculate the random-sample standard error, compare apparent precision, and decide which estimate should inform procurement. They discover why a much larger convenience sample can be more precise about the wrong population.

Where students struggle

Students often equate sample size with quality. Force them to state how units entered the sample before discussing margins of error.

Reading and a quick check

Core textbook chapters on data sources and sampling distributions. Quick check: ask whether doubling sample size halves the standard error and what would be required to halve it approximately.

Applied teaching activity

Run a sampling-design critique: groups choose among stratified, cluster, simple random and convenience approaches for a multi-country customer study and defend cost versus bias trade-offs.

Where it leads

Sampling distributions make it possible to quantify estimation uncertainty through confidence intervals.

Central question: How precisely can the sample estimate the business parameter that matters?

Confidence intervals should be taught as ranges of plausible parameter values generated by a procedure, not as decorative brackets around a point estimate. Students should connect interval width to sample size, variability and confidence level, then translate that precision into managerial usefulness.

What this concept covers

• Point estimates and interval estimates.
• Confidence intervals for means and proportions.
• Margin of error and standard error.
• Sample-size planning.
• Practical interpretation and common misconceptions about confidence levels.

Learning outcomes

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

• Construct intervals for means and proportions.
• Interpret confidence intervals accurately.
• Explain why wider confidence levels produce wider intervals.
• Assess whether an interval is precise enough for a decision.

Supports Course Outcomes 6 and 7.

How to teach it

• Start with a decision threshold so students can see why interval width matters.
• Compare 90%, 95% and 99% intervals using the same data.
• Ask students to plan sample size for a target margin of error.
• Require a plain-English conclusion that names the parameter and business implication.

Case-style example

Brightline Telecom tests a revised onboarding journey with 520 customers. Activation is 61% in the new journey versus a historical benchmark of 56%. Students construct a confidence interval for the new activation proportion and assess whether the plausible range is large enough to justify rollout. One team may prioritise the positive point estimate; another may delay because the interval still includes commercially modest gains. They discover that precision, not just direction, drives a defensible decision.

Where students struggle

Students say there is a 95% probability that the fixed parameter lies in the realised interval. Correct this gently and then shift the conversation to what the interval does tell the decision-maker about precision.

Reading and a quick check

Core textbook chapters on confidence intervals for proportions and means. Quick check: ask which lever, sample size or confidence level, can narrow an interval without changing the observed variability.

Applied teaching activity

Use a sample-size planning workshop where teams trade research cost against required precision for a product-launch metric.

Where it leads

Intervals quantify estimation uncertainty. Hypothesis tests add a structured way to challenge a benchmark or compare evidence against a null claim.

Central question: What does the evidence say against a null hypothesis, and how much should that matter for the business decision?

Hypothesis testing is where business statistics courses most easily become procedural. The teaching aim should be to make the null, alternative, test statistic, p-value, effect size and decision context work together. Students should not be rewarded for writing “reject” without explaining practical significance and uncertainty.

What this concept covers

• Null and alternative hypotheses.
• Type I and Type II errors, power and sample size.
• One-sample tests for means and proportions.
• p-values, effect sizes and confidence intervals.
• Statistical significance versus practical or managerial significance.

Learning outcomes

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

• Formulate testable hypotheses from managerial claims.
• Conduct and interpret common one-sample tests.
• Explain Type I and Type II errors in business terms.
• Avoid treating a p-value threshold as the business decision itself.

Supports Course Outcomes 7 and 8.

How to teach it

• Begin with the business cost of each error before writing H0.
• Use a very large sample with a tiny effect to separate statistical from practical significance.
• Require students to report an effect estimate and interval alongside the p-value.
• Use the phrase “evidence against H0” rather than “proof” or “no effect.”

Case-style example

ClearDesk Software claims a redesigned trial raises conversion above 18%. In a sample of 9,000 prospects, conversion is 18.7%, producing a small p-value. Students run the one-proportion test, estimate the effect in percentage points and calculate the extra annual customers at current traffic. One conclusion is “statistically detectable but commercially weak”; another is “worth scaling because acquisition cost is low.” Students discover that significance does not settle the managerial decision.

Where students struggle

Students confuse “not statistically significant” with “no effect” and “statistically significant” with “important.” Mark the quality of the conclusion more heavily than the mechanical test steps.

Reading and a quick check

Assign Wasserstein, Schirm and Lazar on moving beyond p < 0.05 alongside the textbook hypothesis-testing chapter. Quick check: ask students to rewrite a bad conclusion that says “p = 0.08 proves no difference.”

Applied teaching activity

Use the WSES hypothesis-testing case in Session 7 or 8 so students have to choose tests and connect the result to a marketing decision.

Where it leads

Once students can test one parameter, they can compare multiple groups and categorical relationships.

Central question: Which comparison method matches the outcome, groups and design of the business question?

This concept turns inference into comparison. Students should learn to identify whether they are comparing means, proportions, counts or several group means, and to check whether observations are paired, independent or structured in a way that changes the method.

What this concept covers

• Independent and paired comparisons of means.
• Two-proportion comparisons.
• Chi-square tests of independence for categorical variables.
• One-way ANOVA and the logic of between-group versus within-group variation.
• Multiple comparisons and the risk of data dredging.

Learning outcomes

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

• Select a suitable comparison test from the study design.
• Interpret mean and proportion differences with uncertainty.
• Use chi-square methods for categorical association.
• Explain what ANOVA can and cannot establish about group differences.

Supports Course Outcomes 8 and 9.

How to teach it

• Give a mixed set of business questions and ask students to choose the method before seeing numbers.
• Use a paired before-after dataset to expose the cost of ignoring dependence.
• Treat ANOVA as a variance decomposition story, not only an F statistic.
• Add a multiple-comparison discussion when groups exceed two.

Case-style example

BlueHarbour Fitness tests three retention offers across 360 randomly assigned members. Three-month renewal rates are 54%, 61% and 68%, while average visits differ by less than one visit per month. Students compare categorical renewal with chi-square and continuous visits with one-way ANOVA, then decide which offer to recommend. They discover that the “best” treatment depends on the outcome chosen and that different statistical questions can support different managerial conclusions.

Where students struggle

Students often choose methods from keywords rather than design. Ask first: what is the outcome type, how many groups, and are observations independent or paired?

Reading and a quick check

Core textbook chapters on comparing means, counts and experiments. Quick check: provide four questions and ask only for the correct method plus one assumption.

Applied teaching activity

Run a mini-experiment design workshop where teams choose primary outcomes and analysis before seeing results. This creates a natural bridge to pre-specification and multiple-testing risk.

Where it leads

Group comparisons show whether outcomes differ. Regression generalises the idea by modelling how an outcome changes with one or more predictors.

Central question: How much does one variable help explain or predict another, and what does that relationship not prove?

Regression is often the point where Business Statistics becomes genuinely managerial. Students can move from comparing group averages to estimating relationships, making predictions and testing assumptions. The course should keep correlation, regression, prediction and causation carefully separated.

What this concept covers

• Scatterplots, covariance and correlation.
• Least-squares line, slope and intercept.
• R-squared and residual variation.
• Inference for a regression slope.
• Prediction intervals, extrapolation and the difference between association and causation.

Learning outcomes

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

• Interpret correlation and regression coefficients in business language.
• Evaluate model fit using residuals and R-squared.
• Produce predictions with appropriate uncertainty.
• Identify when a causal interpretation is not justified.

Supports Course Outcomes 9 and 10.

How to teach it

• Start with a scatterplot and ask for a qualitative model before formulas.
• Use a dataset with a strong correlation driven by a third variable.
• Require units in every slope interpretation.
• Compare confidence intervals for the mean response with prediction intervals for a new observation.

Case-style example

UrbanBite has 52 weekly observations linking digital advertising spend to online orders. The fitted model estimates 18 additional orders per £1,000 of spend with R-squared of 0.62, but the largest campaigns occur during holidays. Students estimate the line, inspect residuals, predict orders for a £20,000 campaign and debate whether the slope is causal. They discover that a useful predictive relationship can still be confounded by campaign timing.

Where students struggle

Students over-read R-squared and treat the slope as a causal return on spending. Ask what variable could create the same pattern and what design would be needed to support causation.

Reading and a quick check

Core textbook chapters on correlation, regression and inference for regression. Quick check: ask students to explain why a high R-squared is neither necessary nor sufficient for a good decision model.

Where a simulation fits

The Portfolio Management Simulation fits after covariance, correlation and regression concepts if the lecturer adds the required portfolio-theory bridge. Students use historical data for 25 global companies to estimate expected return, beta, covariance, volatility and portfolio weights before making allocation and rebalancing decisions.

Where it leads

Simple regression opens the door to multiple predictors, diagnostics and the harder question of whether a model is credible enough for decision use.

Central question: How should students build a multivariable model without turning it into an automatic variable-selection exercise?

Multiple regression is where students must combine statistical mechanics with subject-matter judgement. The lecturer should emphasise coefficient interpretation conditional on other variables, omitted-variable concerns, multicollinearity, functional form, residual diagnostics and the difference between predictive and explanatory goals.

What this concept covers

• Multiple regression coefficients and partial relationships.
• Dummy variables and interactions.
• Multicollinearity and omitted-variable bias.
• Residual diagnostics, nonlinearity and heteroskedasticity.
• Model comparison, out-of-sample validation and the limits of stepwise selection.

Learning outcomes

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

• Interpret coefficients conditionally.
• Diagnose common regression problems.
• Build a parsimonious model tied to a business question.
• Separate explanatory claims from prediction performance.

Supports Course Outcomes 10 and 10.

How to teach it

• Use nested models so students see how coefficients change when controls enter.
• Add an interaction that has a clear business interpretation.
• Hold out a test sample and compare in-sample fit with prediction error.
• Make students justify each variable before looking at significance.

Case-style example

SwiftHome Lending models loan default using 8,200 historic accounts. A linear probability model uses income, debt ratio, loan-to-value, employment length and an interaction between debt ratio and income. Adding 18 weak predictors raises in-sample R-squared from 0.21 to 0.24 but worsens holdout error. Students compare models, inspect multicollinearity and recommend a specification. They discover why better in-sample fit can produce worse out-of-sample decisions.

Where students struggle

Students search for stars in software output. Require a model rationale, diagnostics and holdout result before any discussion of p-values.

Reading and a quick check

Use the core textbook multiple-regression chapters and, for advanced cohorts, Athey and Imbens or Mullainathan and Spiess to contrast classical estimation with modern prediction.

Applied teaching activity

Use the Flipkart customer analytics case after this concept. Its spreadsheet-based analysis can support a regression workshop in which students translate coefficients into promotional recommendations and discuss omitted variables.

Where it leads

Regression handles cross-sectional prediction and explanation. Time series introduces dependence over time and forces students to think differently about validation.

Central question: How should students forecast a business quantity when observations are ordered in time and the future can behave differently from the past?

Forecasting is a natural capstone for a Business Statistics course because it makes uncertainty operational. Students should distinguish level, trend, seasonality and noise, use time-respecting validation, compare simple baselines with more elaborate methods and report forecast uncertainty rather than a single number.

What this concept covers

• Time plots, trend and seasonality.
• Moving averages and exponential smoothing.
• Trend regression and basic decomposition.
• Forecast error measures such as MAE, RMSE and MAPE with their limitations.
• Back-testing, rolling validation and prediction intervals.

Learning outcomes

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

• Diagnose basic time-series structure.
• Generate and compare simple forecasts.
• Use holdout periods appropriately.
• Explain why forecast uncertainty grows with the horizon.

Supports Course Outcomes 10 and 10.

How to teach it

• Begin with a naive forecast and require any complex method to beat it.
• Use a rolling-origin validation exercise rather than random train-test splitting.
• Show a promotional spike that should not be mistaken for permanent trend.
• Ask students to report a range and a central forecast to an operations manager.

Case-style example

NorthPier Grocers forecasts weekly demand for a chilled meal across 104 weeks. The latest 26 weeks include a 12% upward trend and strong December peaks. Students compare a seasonal naive forecast, a moving average and trend-plus-seasonal model on the final 13 weeks using MAE and MAPE, then recommend a method for replenishment. They discover that the most complex model is not automatically best and that validation design can change the ranking.

Where students struggle

Students randomly split time-series data, leak future information into model selection or optimise one error measure without considering the operational cost of under- and over-forecasting.

Reading and a quick check

Assign the M5 competition readings to show how forecasting methods are compared at scale and how uncertainty forecasts matter. Quick check: ask why a random train-test split can give an unrealistic estimate of future forecast performance.

Where a simulation fits

The Financial Statement Analysis Simulation can support a lighter trend-and-updating discussion because students reassess a company across five reporting periods. It should not be described as a formal time-series forecasting exercise, but it is useful for showing how new data should update a prior judgement.

Where it leads

The final concept asks students to integrate statistical evidence with decision costs, model risk, ethics and communication.

Central question: How should statistical evidence be converted into a business recommendation when models are imperfect and consequences are asymmetric?

A strong course finishes by making statistics serve judgement. Students should weigh effect size, uncertainty, decision thresholds, false-positive and false-negative costs, model limitations, fairness, reproducibility and communication to non-technical stakeholders. This is also the natural place to address AI-assisted analysis.

What this concept covers

• Decision thresholds, expected consequences and asymmetric loss.
• Sensitivity analysis and model risk.
• Responsible reporting, reproducibility and selective presentation.
• Fairness, subgroup performance and ethical statistical practice.
• Executive communication of estimates, uncertainty and limitations.

Learning outcomes

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

• Translate statistical results into a defensible recommendation.
• Perform basic sensitivity analysis around key assumptions.
• Identify ethical and communication risks in statistical work.
• Defend what the data do and do not support under questioning.

Supports Course Outcomes 10 and 10.

How to teach it

• Give students the same evidence but different decision costs and compare recommendations.
• Require a one-slide executive summary with one chart, one estimate, one uncertainty statement and one limitation.
• Use an ethics vignette involving subgroup performance or omitted results.
• Finish with oral challenge: what evidence would make you change your recommendation?

Case-style example

Cobalt Insurance tests a pricing model on 60,000 policies. Overall error falls 9%, but absolute error is 17% higher for customers in one small regional segment. A rollout saves an estimated £2.4 million annually if aggregate performance holds. Students compare full rollout, monitored rollout and delay, quantify the business case and specify fairness checks. They discover that an average performance gain can coexist with subgroup risk that changes the governance decision.

Where students struggle

Students often think the analyst’s job ends when the software produces a result. Mark the decision rule, limitation statement, evidence quality and response to challenge as explicit outputs.

Reading and a quick check

Use the ASA Ethical Guidelines, the 2023 fairness review and the recent LLM/econometric framework for advanced cohorts. Quick check: ask students to list one statistical, one operational and one ethical reason a model with strong average accuracy might still be unsuitable.

Where a simulation fits

The Portfolio Management Simulation is the stronger capstone if students receive the finance primer: teams must turn expected returns, covariance and risk metrics into allocations, then update decisions as news arrives. The Financial Statement Analysis Simulation is a lower-prerequisite alternative focused on individual evidence, trend interpretation and revision of judgement across reporting periods.

Where it leads

This closes the course by making the central principle explicit: statistical output informs a decision, but it does not make the decision.

10-week course structure

A 10-week version should combine adjacent concepts rather than remove inference, regression or forecasting.

14-week course structure

A 14-week version creates space for experiment design, model diagnostics and a separate simulation/debrief session.

The Financial Statement Analysis Simulation places each student in the role of a Financial Analyst. Students analyse the Income Statement, Balance Sheet and Cash Flow Statement, calculate profitability, liquidity, efficiency and solvency ratios, interpret changes across five reporting periods and form a supported share-price prediction. For a Business Statistics course, the pedagogical value is the visible sequence from calculation to trend interpretation to revision of a judgement as new data arrive.

Why it fits here

• Students can practise descriptive comparison across periods and explain direction, magnitude and uncertainty in a business context.
• The single-player structure provides attributable evidence, which is useful if the course needs an individual component.
• The simulation is beginner accessible, so it can sit early in the course without a finance prerequisite, although a short explanation of the statements and ratios is still good teaching practice.

What students do

• Watch the introductory briefing and review their mission, task checklist and sequence of stages.
• Work with the three primary financial statements and calculate required ratios across the first two years using supplied formulas.
• Classify ratio movements as positive, negative or inconclusive and write rationales for those judgements.
• Move through five evolving analysis rounds with updated financial information and company developments.
• Update selected ratios, identify trends and warning signs, interpret qualitative information and submit revised share-price predictions.
• Use the reflection stages to review correct answers, compare decisions with the cohort and explain when and why their judgement changed.

Recommended course placement

• After Session 2 if used as a descriptive-statistics application.
• Across Sessions 2-3 if the lecturer wants time for a separate debrief on trends, medians and cohort variation.
• As an individual formative exercise before students begin formal inference.

Knowledge students should have before the simulation

• Ability to calculate percentages and rates.
• Basic interpretation of change over time.
• A short lecturer briefing on the three financial statements if the cohort has no accounting background.

Delivery format

Suggested debrief questions

• Which ratio or trend most influenced your prediction, and why?
• When did new information change your view most strongly?
• Did you follow the cohort median or your own evidence?
• Which metric looked important at first but became less useful later?
• What would you need before treating the share-price prediction as a formal valuation rather than an engagement device?

The Portfolio Management Simulation is the stronger advanced fit because students use historical market data across 25 global companies to estimate expected returns, beta, covariance, volatility and portfolio weights, then construct and rebalance a live equity portfolio. The statistical content is real, but the simulation is not a beginner Business Statistics activity: the live page requires CAPM, Sharpe Ratio and mean-variance optimisation. Build a short finance bridge before launch.

Why it fits here

• It turns expected value, variance, covariance and regression-style beta interpretation into allocation choices.
• Students see that the same return can be evaluated differently under different risk objectives.
• Later quarters force teams to update a model-based strategy as new information arrives rather than treating the first analysis as final.

What students do

• Review their Hedge Fund or Pension Fund client mandate and understand whether success is judged primarily through Alpha or Sharpe Ratio.
• Download historical market data for 25 global companies and use the supplied Excel model to estimate expected return, beta, covariance, volatility and portfolio weights.
• Apply CAPM and mean-variance optimisation to develop an initial allocation, including decisions about concentration, cash and leverage where permitted.
• Execute Quarter 1 purchases, then review portfolio performance and new company or market news in Quarters 2-4.
• Buy, sell or rebalance only when the evidence justifies a change in strategy.
• Finish by reviewing holdings, trade history, return, volatility and the relevant role-specific performance measure, then compare teams only within the same mandate.

Recommended course placement

• After Session 9 or 10 once students have covered covariance, correlation, regression interpretation and model diagnostics.
• At Session 12 as a capstone if a 45-60 minute pre-brief covers CAPM, Sharpe Ratio and mean-variance optimisation.
• Across two classes in a 14-week course, with a separate statistics-focused debrief on assumptions and rebalancing.

Knowledge students should have before the simulation

• Expected return and variance.
• Covariance and correlation.
• Basic regression interpretation, especially beta as a sensitivity measure.
• CAPM, Sharpe Ratio and mean-variance optimisation through the added finance primer.

Delivery format

Suggested debrief questions

• Which estimate had the greatest influence on your initial allocation?
• Where did covariance change a decision that individual-stock volatility alone would not have changed?
• Which news item should have changed the model inputs, and which should have been ignored?
• Did your team rebalance because the evidence changed or because recent performance was salient?
• How would your conclusion change under the other client mandate?
• What statistical assumption would you test before repeating the strategy with real capital?

Grading with the Financial Statement Analysis Simulation

The simulation is single-player, so it can produce individual evidence about ratio calculation, trend interpretation and revision of judgement across reporting periods.

• Accuracy of calculations where the task requires a correct numerical result.
• Quality of the rationale used to classify a trend as positive, negative or inconclusive.
• Consistency between evidence and the share-price judgement.
• Ability to explain when and why the student changed their view.
• Post-simulation reflection that identifies one calculation error and one interpretation error.

Grading with the Portfolio Management Simulation

The simulation is team-based. Students analyse historical data, build and rebalance portfolios and are compared within their Hedge Fund or Pension Fund mandate. Team outcomes are useful evidence, but they should not be treated as individual grades by themselves.

• Quality of the statistical rationale behind expected return, covariance, volatility and portfolio-weight decisions.
• Consistency between the client mandate and the chosen risk-return profile.
• Evidence-based rebalancing rather than reaction to recent performance.
• Interpretation of Alpha or Sharpe Ratio in the context of the assigned mandate.
• Individual reflection or oral defence linking one team decision to the student's own reasoning.

The platform records decisions, submissions and comparative outcomes that can support academic judgement. It does not replace moderation, local marking criteria or attributable individual evidence. To reduce free-riding, combine team simulation evidence with a short individual assumptions note, reflection or viva and sample-mark borderline cases during moderation.