Determinants of Daily Sales in a Student-Oriented Food Truck : A Regression Analysis

Sammanfattning

This thesis estimates an empirical demand model for a student-oriented food truck operating atKTHRoyal Institute of Technology in Stockholm, Sweden. Using 85 days of transaction-level datafrom August 2025 to March 2026, combined with daily weather observations and the first-yearINDEK lecture schedule, we identify which factors significantly affect daily revenue and quantifytheir effects through Ordinary Least Squares regression.Model selection proceeds via exhaustive best subset selection combined with forward stepwisesearch across 21 candidate variables, evaluated by the Bayesian Information Criterion (BIC) andAdjusted R2. The BIC-optimal model (k = 9, n = 85, Adj. R2 = 0.634) identifies six statisticallysignificant demand drivers. Non-linear temperature effects (temp2) confirm that warmer autumndays substantially increase foot traffic. Each INDEK first-year lecture overlapping the lunch window (11:00–14:00) adds approximately 417 SEK in daily revenue, capturing a captive-audienceeffect among students bound to campus. Nymble fair days (Armada, D-dagen) generate a surplusof 1,477 SEK relative to a typical day. The promotional campaign during mottagningsveckan —in which bowls were priced at 80 SEK instead of the regular 95 SEK — yields an additional 1,783SEKdespite the lower unit price, as volume effects dominate. Days on which new students receivedfree lunch reduced sales by approximately 1,156 SEK, demonstrating a direct demand cannibalization effect. Finally, each additional day into a study period reduces daily revenue by 17.7 SEK,consistent with declining lecture attendance and the gradual formation of home-cooking habits.Bootstrap validation (1,000 iterations) confirms that the k = 9 model generalises well to unseendata (out-of-bag R2 = 0.453), with an overfit gap of 0.251. All classical OLS assumptions aresatisfied: residuals are homoscedastic (Breusch–Pagan p = 0.551), normally distributed (JarqueBera p = 0.510), non-autocorrelated (Durbin–Watson = 2.118), and the functional form is correctlyspecified (RESET p = 0.898). All Variance Inflation Factors are below 2.5, confirming the absenceof multicollinearity.Aprofitability analysis of the campaign pricing shows that the 80 SEK price generates approximately 367 SEK more gross profit per day than normal pricing, at the cost of serving roughly28 additional bowls. The long-run case for the campaign is strengthened by customer acquisition:students who tried the food during mottagning became regular customers whose lifetime valuesubstantially exceeds the short-run margin sacrifice. The results demonstrate that standard econometric methods can deliver actionable operational insights for small food service operators, evenfrom limited data.Keywords: demand estimation, OLS regression, food truck, weather effects, best subset selection,bootstrap validation, BIC, student-oriented food service, price elasticity, KTH.