Assessing the effects of recent events on Chipotle sales...
Transcript of Assessing the effects of recent events on Chipotle sales...
1Dr. Simon Sheather SAS Day 2016
Assessing the effects of recent events on Chipotle sales revenue
22
InFebruary2005,Imovedfrom
33
HeadoftheDepartmentofStatistics:March1,2005untilFebruary28,2014
44
InFall2007,MS(Statistics)onlinebeganwith20students
55
In2012,TexasA&MStatisticalServicesLPwasformed
66
InFall2013,MS(Analytics)face‐to‐faceandonlinebegan
77
PredictingquarterlyrevenueforChipotleMexicanGrill
• WewishtodevelopatimeseriesmodelforthequarterlyrevenueofChipotleMexicanGrill,inordertoprovideforecastsofquarterlyrevenueforthenexttwoquarters,namelyquarters3and4of2016.WeshallconsiderChipotle’squarterlyrevenuefromthefirstquarterof2004untilthelatestavailableresultsnamely,thesecondquarterof2016.Weshalldevelop2models,namely,
1. Atimeseriesmodel2. Aregressionmodel(withtimeserieserrors)usingthe
followingpredictors1) TotalnumberofexistingChipotlestoresatthebeginningof
eachquarter2) NumberofnewChipotlestoresopenedduringeachquarter
88
Source: http://www.foodsafetynews.com/2015/12/a‐timeline‐of‐chipotles‐five‐outbreaks/#.WAz‐H3nrvbh
99
1010
Regressionmodel
Conclusions:• Evidence of non constant variance in the residuals, we shall consider a logarithmic
transformation of revenue and the predictors.
1111
Regressionmodel
Conclusions:• Evidence of outliers in the residuals• Evidence of autocorrelation in the residuals
1212
Residualsfromregressionmodel
1313
Residualsfromregressionmodel
Conclusions:• We shall consider an AR(5) model for the residuals from the least squares regression
model
1414
Residualsfromregressionmodel
Conclusions:• We shall begin by considering an AR(5) model for the residuals from the least squares
regression model
1515
RegressionmodelwithAR(5)errors
Notice that the coefficient of Log[New Store Openings] is positive and statistically significant
1616
RegressionmodelwithAR(5)errors
1717
Comparisonofregressionmodels
1818
RegressionmodelwithARgapmodelerrors
Conclusions: For every 1% increase in • Existing stores, Chipotle’s quarterly revenue is predicted to increase by 100*(1.011.26554 – 1) = 1.27%• New stores, Chipotle’s quarterly revenue is predicted to increase by 100*(1.010.03651 – 1) = 0.04%
1919
RegressionmodelswithARgapmodelerrors
2020
RegressionmodelwithARgapmodelerrors
2121
Timeseriesmodel
Conclusions:• Quarterly revenue increases over time, as does the variability in the quarterly revenue. • In order to stabilize the variability, we shall consider a logarithmic transformation of revenue.
2222
Timeseriesmodel
Conclusions:• A logarithmic transformation of revenue stabilizes the variability of the series• With an increasing trend in Log[Revenue ($M)] there is a need for differencing
2323
Timeseriesmodel
Conclusions:• Apart from 2013, Log[Revenue ($M)] is lower in the first quarter when compared to the other
three quarters, for which Log[Revenue ($M)] is similar each year. • In other words, there is evidence of seasonality in the log transformed revenue data.
2424
Timeseriesmodel
Conclusions:• The Ljung‐Box Q statistic is statistically significant at lags 1, 7 and 8, with lag 1 being the most
statistically significant. • With parsimony in mind, we begin by considering models just based on lag 1 autoregressive (AR)
and/or moving average (MA) terms.
2525
Timeseriesmodel
Conclusions:• The seasonal ARIMA(0,1,1)(0,1,0)4 model produces the lowest values of AIC (Akaike
Information Criterion) and SBC (Schwartz Bayesian Criterion) and hence is deemed to be the “best” of the four models considered.
2626
Timeseriesmodel
Conclusions:• The seasonal ARIMA(0,1,0)(0,1,0)4 is a valid model.• A number of outliers are evident in the plot of the residuals
2727
Timeseriesmodel
Time Type2014Q1 Level Shift (LS)2013Q2 Level Shift (LS)2016Q1 Additive Outlier (AO)2015Q4 Level Shift (LS)2006Q1 Level Shift (LS)
2828
Timeseriesmodelallowingfor1AOand4LS
2929
EstimatedpercentagechangeinChipotle’squarterlyrevenuedueto4levelshiftsandanadditiveoutlier
Time Type % change
2006Q1 Level Shift (LS) ‐5.7
2013Q2 Level Shift (LS) ‐14.0
2014Q1 Level Shift (LS) 22.9
2015Q4 Level Shift (LS) ‐15.5
2016Q1 Additive Outlier (AO) ‐13.1Chipotle (https://chipotle.com/2015incidents) reports on the following “food safety incidents of 2015”
Norovirus in August and December, 2015 Salmonella in August, 2015 E. coli in October‐November, 2015
3030
Timeseriesmodelallowingfor1AOand4LS
Level shift
Level shift
Additive outlier
3131
Timeseriesmodelallowingfor1AOand4LS
3232
Timeseriesmodelpredictions
3333
Source: http://ir.chipotle.com/phoenix.zhtml?c=194775&p=irol‐newsArticle&ID=2206614
3434
Source: http://ir.chipotle.com/phoenix.zhtml?c=194775&p=irol‐newsArticle&ID=2215726
3535
Analyst’sexpectationsaboutrevenue