0
$\begingroup$

According to this article, calculating elasticity of demand for different models is:

enter image description here

I generate data for 5% reduction in prices with a corresponding 10% increasing in sales: price elasticity = (+10%/-5%)=-2 and get different result coefficient from linear model.

price <- double(length = 10L)
price[1] <- 200
for(i in 2:10)price[i] <- price[i-1]*.95 # price sequence with step =  -5% of last value
sales <- numeric(10)
sales[1] <- 1000
for(i in 2:10)sales[i] <- sales[i-1]*1.1 # sales sequence with step =  +10% of last value
plot(price,sales,type="b")
+.1/-.05 #price elastic = -2
m <- lm(log(sales)~log(price))
coef(m) #price elastic = -1.858141 it's not equal -2

Then I take the same dependency of sales on price plus trend and monthly seasonality and again I get another coefficient!

# generate monthly  time series (length=48 month) with trend, monthly seasons and price correlations (from above)
smpl <- sample(x = 1:10,size = 48,replace = T) # create random index for get price-sales dependies
mnth.sez <- runif(n = 12,min = .8,max = 1.2) #create multiplication seasonality
mnth.sez[c(1,12)] <- c(0.5,1.9) # manual adjust for january and december
mnth.sez

(mnth.ts <- rep(x = 1:12,length.out=length(smpl))) # create monthly sequence
sez.ts <- mnth.sez[mnth.ts] # create seasonality sequence
trend.ts <- seq(from=10000,by=100,length.out = length(smpl)) # create linear trend
ts.num <- seq(length(trend.ts)) # create trend predictor for lm model
sales.ts <- trend.ts*sez.ts+sales[smpl] # create sales with trend, seasonality and price-sales dependies
price.ts <- price[smpl] # create price predictor for lm model
plot(x=ts.num,y=sales.ts,col="black",type = "b")
text(x=ts.num,y=sales.ts,labels = mnth.ts,pos=2,col="blue")
text(x=ts.num,y=sales.ts,labels = round(price.ts),pos=4,col="red")
# View(data.frame(mnth.ts,sez.ts,trend.ts,sales.ts,sale.price=sales[smpl]))
d <- data.frame(sales.ts,ts.num,price.ts,mnth.ts=as.factor(mnth.ts))
d.mat <- as.data.frame(model.matrix(~.,d)[,-1])
m2 <- lm(log(sales.ts)~log(price.ts)+ts.num+mnth.ts2+mnth.ts3+mnth.ts4+mnth.ts5+mnth.ts6+mnth.ts7+mnth.ts8+mnth.ts9+mnth.ts10+mnth.ts11+mnth.ts12,data = d.mat)
summary(m2)
coefficients(m2) # -0.23 it's not equal  -2 and -1.858141!!!
lines(x = ts.num,y=exp(fitted(m2)),col="green")

Why is it incorrect, and what is the true strategy for price elasticity extraction in trend-seasonality problems?

$\endgroup$
0
$\begingroup$

If $P2=P1*0.95$ then $P3=P2*0.95=P1*(0.95)^2$. So if $P1=100$ then $P2=95$ and $P3=90.25$ which isn't a 5% price step. I think your data generating process is incorrect which leads to a price elasticity from the model that doesn't match what you think it ought to be ($\frac{0.1}{-0.05}=-2$). As for the latter question, I do not have an answer for the price elasticity extraction in trend-seasonality problem.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.