Loading data…
BUS173 topic: descriptive statistics and index numbersHeadline results from each part of the course, all from the same WFP price data.
Staple Fisher index vs official food CPI
2016 = 100, monthlyBUS173 topic: index numbers (Laspeyres, Paasche, Fisher, chain)Prices by division, the four ways of turning them into one index, and how the result compares with the official food CPI.
Commodity price explorer
How to read this: each thin line is the median price across the WFP markets in one of the six older divisions. The thick line is the national price, the mean of the division medians. The four long-basket staples start in the mid-2000s; the other items start in 2024.
Four index formulas, long basket
2016 = 100Weights
Base quantities per person per month| Basket | Item | q0 | p0, BDT | Share | Source |
|---|
How to read this: q0 is kg (litres for oil, pieces for eggs) per person per month, from BBS HIES Table 5.2 daily grams × 30.4. Share is p0 × q0 over the basket total. The long basket uses HIES 2016; the 11-item short basket uses HIES 2022.
Year-on-year inflation, monthly
%, staple Fisher index vs food CPIAnnual average inflation
%, full calendar yearsStaple inflation ranges from −28.0% (2009) to +30.2% (2010). The food CPI stays between 3% and 13% because it averages over many more items, including ones whose prices move slowly.
BUS173 topic: hypothesis tests, ANOVA, chi-square and contingency tablesEvery test with a p-value in notebooks 02 and 03, with Holm-corrected p-values, and the tables behind the main findings.
Hypothesis test results
| Test | Data | n | Statistic | p | Holm p | Effect | Conclusion |
|---|
How to read this: Holm p adjusts for running many tests in the same section. A green edge marks a test still significant at 5% after that correction. Monthly prices are autocorrelated, so where both exist, trust the Newey-West and permutation versions over the plain t-test and chi-square.
Is there a Ramadan price rise?
Ramadan month vs two months before, net of drift, 2007-2026Division price levels, 2025
% above or below the national pricePrice spikes vs floods, rain and Ramadan
2 × 2 tables, spike = monthly change above the item's 90th percentile| Event | Event: spike / no spike | Other: spike / no spike | Spike rate, event vs other | Odds ratio (95% CI) | Fisher p | Permutation p |
|---|
Are monthly price changes normal?
Monthly % change, national price, 2007-2026| Item | n | SD, % | Skew | Excess kurtosis | Shapiro-Wilk p | Jarque-Bera p | Fitted t df | AIC gain for t | Beyond 3 SD |
|---|
How to read this: a normal distribution has excess kurtosis 0 and puts 0.27% of months beyond 3 SD. All four staples have fatter tails, every normality test rejects, and a t distribution with few degrees of freedom fits better (positive AIC gain). The QQ plots are in notebook 03, figure 16.
BUS173 topic: correlation and regressionHow much of a world price change reaches Bangladeshi shelves, and why regressions on price levels mislead.
Correlation of monthly % changes
Pearson rHow to read this: only palm oil moves with its world price month to month (r = 0.43). Domestic rice and the Thai rice price barely move together (r = 0.06).
Spurious correlation demo
Same pairs, levels vs monthly changesLong-run pass-through
Sum of lag 0-3 coefficients, 95% Newey-West CIOut-of-sample check
Fit 2006-2019, test 2020-2025| Item | Test months | RMSE, model | RMSE, mean | Ratio | OOS R² |
|---|
BUS173 topic: time series, seasonal indices and forecastingTrend, seasonality and volatility of the staples, and a six-month forecast checked against a rolling back-test.
Seasonal indices
Ratio to moving average, 100 = average monthTrend
Six-month forecast
Back-test
13 rolling origins, Jul 2022 to Jul 2025| Series | Method | MAPE h1-6 | MAPE h7-12 | MASE h1-6 | MASE h7-12 |
|---|
How to read this: MAPE is the mean absolute % error. MASE divides the error by that of a one-step naive forecast in the training data, so below 1 beats it. Up to six months ahead SARIMA and ETS are almost tied: SARIMA is slightly better for rice and ETS for the staple index. At 7-12 months the seasonal naive forecast does better for both.
Volatility ranking
SD of monthly % change since Oct 2024Rolling 12-month volatility
SD of monthly % change, long basketBUS173 topic: probability, Bayes' rule and decision analysisA buyer needs 1,000 L of palm oil in each of the next three months. Buy all 3,000 L now and store it, or buy each month at that month's price? The answer depends on the chance of a price spike next month.
Assumptions
adjustable6%18%
0%1.5%
0.5×1.5×
Signal: world palm oil in taka up more than
This month the signal is
Base case: 11.86% all-bank lending rate (Bangladesh Bank, Jun 2026), 0.5% a month storage and loss (assumed), 5% signal. Price now: .
Expected cost of each action
BDT per litre, against P(spike next month)Bayes' rule table
| State | Prior | Likelihood | Joint | Posterior |
|---|
Payoff table
Cost in BDT of covering 3 × 1,000 L| State | Forward buy | Buy monthly | Forward saves |
|---|
Back-test, 2020 to 2025
Base-case settings, 5% signal| Strategy | Total cost, BDT m | vs always monthly | BDT per litre |
|---|
How often does palm oil spike?
Beta-binomial posterior of the monthly spike rate| Stage | Beta(a, b) | Mean | 95% credible interval |
|---|