Statistics for market analysis

#1
Hello there

First of all I am not a pure statistician, however I worked with Weibull stats during my PhD.

Now that I am working in marketing, I realized that "customers' orders" can be modeled with Weibull (someone can argue: what can't be modeled with Weibull?!?... Weibull is life!).

In other words, I consider the first day of fiscal year as "0", and I plot Weibull reliability vs number of days until I get an order from a customer.

Ex: Jan-2nd --> day 1 --> order 1
Jan-10th--> day 9--> order 1
Feb-2nd--> day 32--> order 5
... and so on
the number of orders I get on specific day is counted as grouping (or weight)

Question 1: Anyone here has any experience/advice to share about the approach and/or methodology? Maybe something better than Weibull?

Question 2:
I want to build a tool to predict market evolution based on past experiences: I have past years data and, except some similarities, every year has its own specifications. As I am feeding current year data daily/monthly, is there a way to simulate, with some confidence, if this current year will behave in a same way as a previous year?

Thank you for your help
 

Miner

TS Contributor
#2
Q1: This is straight forward reliability/survival analysis for events in time. There are many distribution other than Weibull that may also be used as well as non parametric methods.

Q2: You might try reliability growth modeling to predict the current year from prior years. Then plot the actuals against the prediction line as they occur. If the assumptions are incorrect the actuals will begin to diverge from the prediction line.
 
#3
Miner,

Thank you for your reply.

Concerning Weibull statistics, it is true that it is mainly used for reliability analysis... However, its applications can be extent beyond this specific field

Concerning "reliability growth modeling", I am not familir with it, can you please tell me more about it?

Thank you
 

Miner

TS Contributor
#4
I have no disagreement on its flexibility outside reliability. The ability of the Weibull distribution to model such a wide variety of shapes is why it is so widely used. My real point was that there are also other distributions that may be used to model events in time (see attachment), so don't restrict yourself to the Weibull only.

Regarding reliability growth, this is a method for determining early on whether the initial results are meeting the target goals, but may also be used to determine whether baseline statistics are remaining constant or changing. The two most common models are the Duane and the Crow-AMSAA models.
 

Attachments

#5
Miner,

Thank you so much for the attachement. I will try to dig into these distributions.
As I said, I am not a statistician, so I came here precisely to have other points of view. I am very glad I could cross your way!

Concerning Weibull, I thought this distribution quite elegante:
- if my customers rush their orders --> beta factor <1
- if my customers put their orders randomly --> beta factor =1
- if the population of my customers put orders with an intrinsic pattern (for example: seasonal orders) --> beta factor>1

or maybe my reasoning is wrong?
 

Miner

TS Contributor
#6
I would phrase it differently:
- Beta < 1: Events are occurring at a decreasing rate (Orders are decreasing)
- Beta = 1: Events are occurring at a constant rate (Customers are putting there orders in randomly)
- Beta > 1: Events are occurring at an increasing rate (Orders are increasing)