Note on Consumer Behavior - MIT
web.mit.edu/hauser/.../Hauser%20...
Massachusetts Institute of Technology
by JR Hauser - Related articles
John R. Hauser. In a classic paper on the ... better understanding of the various ―irrational‖ influences on pur- chase decisions and ... consumers to self-[PDF]consumer preference axioms: behavioral postulates for ... - MIT
web.mit.edu/hauser/.../HauserAxio...
Massachusetts Institute of Technology
choice theory, and consumer behavior to develop five basic axioms or postulates of stochastic choice ... 1332 JOHN R. HAUSER. 2. Existing ... Examining the various approaches to understanding or predicting consumer pre- ferences, we see ...
A regression model for predicting optimal purchase timing for
airline tickets
William Groves and Maria Gini
Department of Computer Science and Engineering, University of Minnesota
{groves, gini}@cs.umn.edu
Abstract
Optimal timing for airline ticket purchasing from the consumer’s perspective is challenging principally
because buyers have insufficient information for reasoning about future price movements. This paper
presents a model for computing expected future prices and reasoning about the risk of price changes.
The proposed model is used to predict the future expected minimum price of all available flights on specific
routes and dates based on a corpus of historical price quotes. Also, we apply our model to predict prices
of flights with specific desirable properties such as flights from a specific airline, non-stop only flights, or
multi-segment flights. By comparing models with different target properties, buyers can determine the
likely cost of their preferences. We present the expected costs of various preferences for two high-volume
routes. Performance of the prediction models presented is achieved by including instances of time-delayed
features, by imposing a class hierarchy among the raw features based on feature similarity, and by pruning
the classes of features used in prediction based on in-situ performance. Our results show that purchase
policy guidance using these models can lower the average cost of purchases in the 2 month period prior
to a desired departure. The proposed method compares favorably with a deployed commercial web site
providing similar purchase policy recommendations.
1 Introduction
Adversarial risk in the airline ticket domain exists in two contexts: the adversarial relationship between buyers
and sellers, and the competitive relationships that exist between multiple airlines providing the equivalent
service. Buyers are often seeking the lowest price on their travel, while sellers are seeking to keep overall
revenue as high as possible to maximize profit. Simultaneously, each seller must consider the price movements
of its competitors to ensure that its prices remain sufficiently competitive to achieve sufficient (but not too
high) demand. It is impossible to effectively address the problem of optimizing decision making from the
buyer’s point of view without also considering both types of adversarial relationships.
Sellers (airlines) make significant long term investments in fixed infrastructure (airports, repair facilities),
hardware (planes), and route contracts. The specific details of these long term decisions are intended to
roughly match expected demand but often do not match exactly. Dynamic setting of prices is the mechanism
that airlines use to increase the matching between their individual supply and demand profile in order to
attain the greatest revenue.
A central challenge in the airline ticket purchasing domain is the information asymmetry that exists
between buyers and sellers. Airlines have the ability to mine significant databases of historical sales data to
develop models for expected future demand for each flight. Demand for a specific flight is likely to vary over
time and will also vary based on the pricing strategy adopted by the airline. For buyers, it is generally best
to buy far in advance of a flight’s departure because the prices tend to increase dramatically as the departure
date approaches. But, airlines often violate this principle and adjust prices downward to increase sales.
We make two novel contributions in this work: (1) a method of automated optimal feature set generation
from the data that leverages a hierarchicalization of the available features to efficiently compute a feature set
is proposed; (2) the addition of time-delayed observations to the feature vector fed to the machine learning
1
algorithm is performed. This allows anticipation of trends and more complex relationships between variables.
For instance, we address pricing behaviors up to and beyond 60 days prior to departure, and we consider
purchasing a flight on any airline for a specific date and city pair (previous work only considers the cost of
a specific pair of flight numbers from two specific airlines).
These ideas are then experimentally applied to prediction in the real-world airline ticket purchase domain.
This paper presents models that also accommodate preferences of passengers about the number of stops in
the itinerary or the specific airline to use. We believe this prediction task is both a more difficult task and
generates models that are more useful for actual airline passengers.
Airline customer experience modeling: A
Simplified Version of Conjoint Analysis
Airline flight packages can come in many versions or with
many features. Each feature is costly to
include, and in general, we assume that including the feature will be
profitable if the consumers’ willingness to pay (WTP) for that feature exceeds
the cost of including that feature by a comfortable margin.
To further simplify the algorithms modeling, one can assume
that all customers have the same preferences – the same WTP for each feature,
and this assumption does not hold in real markets, we will therefore have to
consider preferences either by segment or we have to model some preference or
utility distribution across all potential consumers. We do this by estimating a
conjoint model for each consumer or by estimating how WTP varies across
consumers.
1. WTP Attributes
(profitable to airlines if the consumers are willing to pay, WTP)
In addition to fly
ticket price range from $250 to $350, we assume there are only three features
of interest, or “perks” customer appreciate and are willing to pay
Free drinks and
snacks---either free drinks only, or free drinks plus free snacks.
Extra legroom ---seats either with a little more
leg room or with quite much more leg room
Entertainment
---either free news channels or free news channels plus free movie
channels.
With four WTP attributes varying (3 “perks” features plus
price), at two levels each, there are 2x2x2x2 = 2^4 = 16 possible combinations
or profiles. And importantly, we assume we will be able to create pictures/or
short videos of each of the sixteen GPSs and have consumers evaluate all
sixteen GPS “profiles.” We have consumers evaluate or “experience” all sixteen
flight packages or “profiles.” Customers will rate each potential flight
package on a 100-point scale where 100 means most preferred. We make sure that
consumers understood the features and that the task were realistic.
For a single customer, we will have data as in the following
table. The first column indicates the
customer’s preference for a particular combination or “profile” of feature and
price. The next four columns indicate whether or not the rated fight package has
that feature-price combination. A ‘1’ indicates the feature is at its “high” (“perceived”
by customer) level, e.g., free drinks only rather than free drinks plus free
snacks, and a ‘0’ indicates a feature is at its “low” (“perceived” by customer)
level, e.g., free drinks plus free snacks rather than free drinks only. And in terms of customer’s choice, the data
(‘4’) indicate that consumer prefers least a package of ‘free drinks only, a
little more leg room, and free news channel only and priced at $350. The data
suggest (‘99’) that the same consumer prefers most a flight package of free
drinks plus free snacks, quite more leg room, free news channel plus movie
channel and priced at $250.
The goal of conjoint analysis is to determine how much each
feature contributes to overall preference. This contribution is called the
“partworth” of the feature. In this simplified conjoint analysis, we can use
ordinary leastsquares (OLS) regression with the regression coefficients
produced as the partworths . For example, the partworth of free drinks only
(vs. free drinks plus free snacks) is 9.6 indicating that the customer gets 9.6
“utils (utility)” if free snack is also provided in addition to free drinks.
Similarly, the regression estimates that the customer gets 40.6 “utils
(utility)” if the price is reduced from $350 to $250.
With this regression we compute the customer’s willingness
to pay (WTP) for each feature. Because the consumer gets 40.6 “utils” when the
price is reduced by $100 ($350 reduced to $250), the value of each “util” is
about $2.46, which we obtain by comparing the difference in price to the
difference in the price-partworths: $100/40.6. We can compute the WTP for other
features as well.
there are inevitable measurement errors when the customer
provides his or her preferences on the 16 flight packages. And those measurement
errors will translate themselves into uncertainty in the estimates of the
partworths as indicated by their standard errors. We assume those errors will have a normalized
distribution with mean of “zero” if we ask enough consumers to complete the conjoint
analysis exercise as we designed, we could therefore gain greater statistical power and obtain
estimates of the partworths that are more accurate.
[PDF]Note on Conjoint Analysis - MIT
www.mit.edu/.../NoteonConjointA...
Massachusetts Institute of Technology
Loading...
by JR Hauser - Cited by 10 - Related articles
A '1' indicates the feature is at its “high” level, e.g., 10 feet rather than 50 feet, and a '0' ... The goal of conjoint analysis is to determine how much each feature.[PDF]Note on Conjoint Analysis - MIT
www.mit.edu/.../Note%20on%20C...
conjoint analysis task asks consumers to rate each potential GPS on a ... An entry of '1' indicates the feature is at its “high” level, e.g., 5 feet rather than 50 feet,.
Massachusetts Institute of Technology
Loading...
[PDF]Interpreting the Results of Conjoint Analysis - Sawtooth Software
https://www.sawtoothsoftware.com/.../interpca.pdf
Conjoint analysis provides various outputs for analysis, including part-worth util- ... 30 mpg received a negative utility value, but this does not mean that 30 mpg was .... make worse) a product's overall preference by changing its attribute levels one at ... reduce its features and capabilities, we can observe a loss in relative ...
Sawtooth Software
Loading...
[PDF]9 Things Clients Get Wrong about Conjoint Analysis - Google Research
research.google.com/pubs/archive/41886.pdf
Google
Loading...
by C Chapman - Related articles
This paper reflects on observations from over 100 conjoint analysis projects across the ... However, the successes of CA also raise clients' expectations to levels that can be excessively ... indicates a good feature while negative part worths indicate bad ones. .... 10 (on the X axis) and high utilities for features 2, 5, and 9.[DOC]Conjoint Analysis.docx - NMSU College of Business
business.nmsu.edu/.../Conjoint%20Analysis....
Conjoint analysis isn't just one approach; it's multiple approaches for optimizing ... can indicate what people are willing to pay for different levels of those features. .... Utility scores are additive, so the soup with the highest value to customers ...
New Mexico State University
Loading...
[PDF]A Brief Explanation of the Types of Conjoint Analysis - Qualtrics
https://www.qualtrics.com/wp.../09/ConjointAnalysisExp.pdf
Conjoint analysis is the optimal market research approach for measuring the ... Each product profile represents a part of a fractional factorial experimental ... targets the respondent's most preferred feature and levels, thereby making the ... Although the approach is different, the outcome is still the same in that it produces high ...
Qualtrics
Loading...
Conjoint Analysis Theory - Qualtrics Support
https://www.qualtrics.com/support/.../conjoint-analysis-theory/
Segmentation Analysis: The matrix of respondent by attribute-level .... no interest in purchasing and 10 indicates extremely high interest in purchasing.
Qualtrics
Loading...
Brand Strength: Building and Testing Models Based on Experiential ...
https://books.google.com/books?isbn=332281629X
Martin Walser - 2012 - Business & Economics
... by a specific combination of product/service features in conjoint-analysis 5 can be ... ranging from low, medium to high levels (.78, 1.00, 1.22 index levels) Brands 4 ... The model trimmed of these paths can be depicted as indicated below: ...[PDF]Conjoint and Discrete Choice Designs for Pricing Research - Usc
www-bcf.usc.edu/~tellis/conjoint.pdf
University of Southern California
Loading...
by V Kanetkar - Cited by 4 - Related articles
discrete choice is a useful tool, whether conjoint analysis is able to portray ... a product into its attributes and subsequent valuation of the utility of each .... allows us to assess a consumer's willingness to trade off one feature for another. ..... First[PDF]Feature Fatigue: When Product Capabilities Become Too Much of a ...
https://www.rhsmith.umd.edu/.../Feat...
Robert H. Smith School of Business
Loading...
by DV THOMPSON - 2005 - Cited by 408 - Related articles
significantly, empirical evidence indicates that consumers ... as conjoint analysis or discrete choice analysis, model each ... the utility of a product is based on its potential benefits to .... the high level included the 21 most important features.
No comments:
Post a Comment