Faculty of Economics and Business
MSc in Economics - Master Thesis
Consumer behavior in dynamic pricing models
The effect of behavioral biases on delaying purchase behaviorStudent:
Noam Goldman (10599371)
Supervisor: Prof. Joep Sonnemans
Abstract
The increasing use of dynamic pricing models in various markets and products places consumers in situations which involve a relatively high degree of uncertainty. While trying to make the best purchase decision, consumers may be subject to different behavioral biases which may have an effect on their decisions. Using a controlled experiment, I examine the decision behavior of subjects in a particular situation of purchasing a product under a dynamic prices scenario. Delaying purchase behavior of consumers was tested using different levels of price and valuation uncertainty, which involved a risk of losing and ambiguous probabilities. The results show that only when uncertainty becomes very high, loss-averse subjects will have higher tendency to postpone purchases, possibly until all elements of the situation will be clarified and certain.
1
Introduction
The use of dynamic pricing models, when the price of products changes over time, is very common in some industries. Such models and price schemes are particularly relevant nowadays when internet usage is increasing and the costs of changing prices are redundant. However, dynamic prices expose consumers to situations of decision making under uncertainty.
The uncertainty of this situation can appear from different directions. For example, the price of the product may change, the availability of the product may become uncertain in the future and the consumer’s valuation of the product may change as well.
Consumers who arrive to the market in some point of time have few options - to buy the product immediately, to wait and buy it later or not purchase it at all. If consumers will decide to wait, they might get a better deal if the price will decrease, however, they also face the risk that the product will no longer be available, or a possible option that the price will eventually increase.
Focusing on the demand side, this situation may raise several questions. Why do consumers delay purchases of products with dynamic prices? Is it because consumers learn from price history and anticipate the next period price? Which uncertainty takes the most significant part in the decision process? Which behavioral biases are involved in the consumer decision process?
According to past studies, consumers delaying behavior can be a result of strategic thinking or some behavioral biases inherited in the decision making process (Shen and Su, 2007; Su, 2009). However, it is unclear which of the two is the main drive for this behavior. Do consumers wait in hope to get a better deal or do they wait just to avoid further losses? Also, consumers might act in both ways at the same time. For example, waiting for a lower price while in the same time expecting that their valuation of the product will become more accurate, so they can make a better evaluation of the product.
The question of what determines consumer behavior while deciding about purchase of products is an interesting question due to several reasons. First, this information is important for firms who set the prices in order to in-crease their revenues. With current technology, firms can monitor customers’ choices and search behavior and use it in order to set profit-maximizing prices. They can find out how many times did a certain consumer search for a product in their website, and in which price did he or she decide to eventually buy their product.
Moreover, in a research perspective, examining this question from the consumers’ side may provide additional evidence for decision making process under uncertainty, and the behavioral anomalies that are involved in this process. Also, it can be used to learn more about how consumers evaluate a
purchase situation, and which important factors are they taking into account when making their decisions.
In order to test those questions, I conducted a lab experiment in which participants were given a decision making problem imitating a situation of buying a product with constantly changing prices. The experiment allowed me to observe the behavior and the path of decisions. In addition to the price changes, the experimental design contained a possible change in the valuation of the consumers. This was done in order to increase the uncer-tainty and to simulate a more real life dilemma level.
The results of this experiment imply that consumers are trying to be strategic, and that their actions cannot be characterized as strictly myopic. Thus, they are less likely to buy a product now if they think they could have had a better deal in the future, even if their current payoff from the deal is positive. Consumers are able to anticipate the next period price using the available information given to them, and make a decision according to their expectations. Moreover, the results show that increased uncertainty may cause the consumer to delay purchase more. However, it appears that uncertainty should be extremely high in order for the subject to decide to wait, and a lower level of uncertainty in valuation will not make a significant difference in behavior.
The rest of this paper is organized as follows. In the next section, the main literature about dynamic pricing models and consumers behavior will be reviewed. Section 3 will discuss the research method of this study, the experiment. Section 4 will present a theoretical decision rule model. Section 5 will present the results of the experiment. Section 6 will summarize the main findings from the results and will provide possible explanations. Sec-tion 7 will bring final conclusions with the main implicaSec-tions of this study and suggestions for future research.
2
Literature review
2.1 Dynamic pricing models
The basic competitive market equilibrium is defined by one price for all units of demand. However, there are much evidences for notable level of price variations outside the basic models (Shen and Su, 2007). Clearly, buyers can have different buying behaviors and patterns. For example, their timing of purchase and their flexibility may vary. Firms can use this information to offer each of them different conditions and terms of trade, usually in the shape of different prices, which might increase the firms’ profits (Talluri and Ryzin, 2006). Therefore, if firms are able to identify the buying behavior of each customer and distinguish between them, we should expect to observe price divergence in the market (Shen and Su, 2007).
are using dynamic prices. This is because the practice of it is mainly useful for businesses with high kickoff costs, limited capacity, finite sale period and price-sensitive and stochastic demand (Bitran and Caldentey, 2003). The pricing policies and methods used in the market can fit into two categories: posted-price mechanisms and price-discovery mechanisms (Elmaghraby and Keskinocak, 2003). In this study, the main focus will be on dynamic posted-prices mechanisms, namely ”take-it-or-leave-it” selling methods. Eventually, using dynamic pricing allows the seller to adjust the prices throughout the selling period in response to the realized demand over time, amount of avail-able products and time of sale (Elmaghraby and Keskinocak, 2003; Gallego and Ryzin, 1994).
Identifying the demand characteristics is not an easy task for firms. Same as the variations in prices, demand can also vary over time. Changes in de-mand are especially relevant for the kind of products that use dynamic pric-ing, as they usually involve ”pay-now-consume-later” methods (Su, 2009). As for the identification of the demand, early literature about dynamic pric-ing models and Revenue Management (RM) considered demand to be mainly myopic.
Myopic consumers will always purchase the product if its price is lower than their valuation of it, or less then their willingness to pay (Talluri and Ryzin, 2006). Namely, they make a one-time buying decision when first encounter the product upon their arrival to the market (Shen and Su, 2007). However, according to more recent literature, it seems like not all con-sumers are making their purchase decisions according to this rational rule. Recent research in this field began focusing on modeling different types of consumer behavior and by that implementing a more realistic approach to the firm maximization problem. Findings from this new stream of litera-ture have significant implications on firms’ behavior, as it implies that firms should integrate other types of consumer behavior into their own strategy. As a result, demand is no longer regarded as exogenous and myopic, which further complicates the dynamic pricing models.
2.2 The strategic consumer - learn and form price expecta-tions
One behavior that is important to take into consideration is delaying pur-chase even when consumers’ valuation is higher than the current price of the product. According to Shen and Su (2007), consumers are constantly evaluating other purchase options and make decisions. The authors refer to the practice of delaying purchases to some other point in the future as inter-temporal substitution. There are few possible explanations for con-sumers’ delay in purchase for a product with dynamic prices that are found in the literature. The vast majority of studies consider consumers to be ”strategic”.
Strategic consumers are aware of the dynamic path of prices, and there-fore may time their purchase in order to acquire higher utility (Levin et al., 2009; Liu and Cooper, 2014). Also, consumers learn about end-of-period dis-counts so they may decide to wait and buy the product later, when its price will be reduced (Ovchinnikov and Milner, 2012). It seems like consumers, or specialized third parties, are sophisticated enough and are able to learn the firms’ strategies regarding price changes and availability capacity (Levin et al., 2009).
Anderson and Wilson (2003) focus on strategic consumer behavior in the air-travel industry, which is known to be characterized by dynamic prices. The authors argue that consumers are able to calculate the probability that a cheaper fare will be reopened in the future. They conclude that if this probability is high enough, the consumer will decide to delay purchasing a ticket. However, the authors condition the ability to calculate this probabil-ity on the consumers’ knowledge of the price-time curve for an airline seat. It can be argued that most consumers are not aware of this price curve, which implies that most of them are not necessarily acting strategically. Moreover, it may be not that simple to calculate this probability.
According to Yuan and Han (2011), consumers form price expectations based on past historic prices, and following their expectations they are en-gaged in search behavior. In their analysis, the authors mainly define search behavior as searching for the same product via other seller. However, one can also learn from their model if we define the search behavior as searching for the next price of the same particular seller. This consumer behavior is practically delaying a purchase. Although the authors do not use the term ”strategic” consumers, they assume that buyers search for other options in order to gain higher utility compared to what they have now. Their the-oretical and experimental findings show that the higher the current period price, the more the buyers tend to search. However, if the expected price is high, the buyers will search less. We may understand from their findings that there are two contradicting powers on the consumer decision to search. However, which one of these powers has a stronger effect on the decision of the buyer remains unclear, and is a legitimate question especially for formation of firms’ strategy.
A study by Chen and Schwartz (2008) may provide a partial answer to the question above. This study also used an experimental method to investigate how price patterns affect consumers’ propensity to book a hotel room. They find that consumers learn from the price pattern and form expectation regarding the next price. Consequently, consumers’ tendency to book is affected by these patterns. Interestingly, Chen and Schwartz also introduce the ”availability” factor, as an important component in the nature of dynamic pricing. Thus, the subjects in their experiment are facing a contradicting decision - to wait for a better price, but while waiting, facing the risk of a sell-out. In their experiment, all subjects observed the same
current price, however they used four different price patterns leading to this price. In my opinion, the patterns that they used might be considered to be too obvious to the subjects. Hence, it can be suggested that their price patterns are not realistic, and therefore it is hard to generalize the results of their study to a real situation of hotel room booking. Also, introducing more fluctuations in price patterns may induce more accurate reaction of the subjects, and as a result, different effects.
A different point of view about the strategic consumer is shown in Nasiry and Popescu (2009). The authors provide a theoretical model where the con-sumers are strategic and ”emotional rational”, as they call it. The concon-sumers are anticipating two objects: the firms’ pricing policies and their own feeling of regret. According to Nasiry and Popescu, because the consumers antici-pate the regret, they practically behave in a rational way. The consumers’ regret can take two directions. First, when buying, the consumer might feel regret about a wrong purchase, and second, if decided to not buy, the con-sumer might also feel regret due to a feel of missing purchase opportunity. Important to note, the authors argument is built on the advance selling policies, namely when consumers are facing valuation uncertainty because of differences in time between purchase and consumption.
2.3 Bounded rational consumers
It is true that nowadays, when information availability has increased sig-nificantly compared to previous decades, consumers are able to learn more about firms’ strategies and become more ”strategic” themselves. However, consumers may also be considered as boundedly rational, when their deci-sions are not necessarily a result of strategic thinking. Making a decision is complicated and most individuals tend to rely on few heuristics rules in order to make an easier decision process, which may lead them to somewhat irrational choices (Tversky and Kahneman, 1974). Moreover, often prices are too unstable and difficult to predict, thus the presence of strategic con-sumers cannot be assumed a priori (Li et al., 2011). If concon-sumers are not strategic, waiting behavior may results from different behavioral biases and other psychological factors (Popescu and Wu, 2007; Su, 2009).
Li et al. (2011) investigated the extent to which consumers are really ”strategic”, and questioned their presence in the real market of dynamic pricing. Based on data from the air-travel industry, they find that only 5-19 percent of the consumers are acting in a strategic way. Moreover, they find that consumers are more likely to act in a strategic way if they arrive early to the market (so they have time to wait), or if they arrive at the last minute when availability factors are already on stake. Even though the authors find evidence for strategic behavior, one can still argue that 5-19 percent of strategic consumers are relatively low and appear only under specific conditions. Therefore, the overall effect of this behavior is
still remain unclear.
Popescu and Wu (2007) model the consumers as boundedly rational. According to their model, consumers have memory and therefore recall the former prices as strategic consumers do. However, according to Popescu and Wu, consumers are not strategic and are inclined to human decision making biases which they are not aware of. When consumers revisit firms, it raises memory effects, since the demand is affected by previous prices, which the authors describe as forming reference points. Therefore, delaying purchase decisions result from price assessment compared to certain anchor price, that serves as a reference point for discounts or surcharges. In their study, the reference price is updated in each period and is affected by past prices.
Another paper by Nasiry and Popescu (2011) argues that consumers perceive prices as discounts or surcharges relative to some ”memory base reference point”, as in Popescu and Wu (2007). However, in this paper the proposed memory base reference prices are based on the ”peakend rule” -consumers mainly remember the lowest and the most recent prices. Thus, the reference price is a weighted average of the two. The formation of the reference price this way shows that consumers’ loss aversion is strongly connected to the path of prices which the consumers were exposed to (Nasiry and Popescu, 2011).
An important question which rises following these two papers is which reference point is most suitable in the situation of dynamic pricing. For example, forming the reference point can be seen from another direction. Consumers might remember the highest price and to anchor it more than the lowest price. Also, both of these studies focus only on situations where consumers revisit the company and remember its previous prices. However, it can be argued that waiting behavior can be observed also when consumers encounter the firm and its products’ prices for the first time, and that the reference prices are ”borrowed” from another experience.
Su (2009) forms a theoretical model of consumer waiting behavior in the context of dynamic pricing. The author explains the reason for waiting as ”inertia”, the innate tendency to avoid making any purchase. Interestingly, the model which he developed describes ”how consumers may behave and not how they should behave”. For simplicity, Su uses only two periods model where consumers can buy or wait in the first period, and in the second pe-riod they buy the product if the price is below their valuation. Also, the consumer is defined to be only one of two types, ”high valuation” or ”low valuation” consumers. Those simplifications make it easier to formulate the model, however, I believe it also makes it harder to evaluate the real market action of it. Compared with other studies, Su does not claim that consumers cannot be strategic, but together with it they may still demonstrate waiting behavior as a result of other behavioral factors. According to Su’s find-ings, the inertia is related to few of the main behavioral regularities known
in behavioral economic literature: loss-aversion, probability weighting, and hyperbolic time preferences. Further elaboration of his model may reveal more biases which are relevant for the discussion about consumer behavior in dynamic pricing models.
2.4 Behavioral biases related to consumer behavior
Loss aversion is a behavioral bias where sensitivity to losses and disadvan-tages is higher compared to perceived benefits from gains. Therefore, the possibility of losses may affect preferences and behavior more than gains (Tversky and Kahneman, 1991). Su (2009) argue that when consumers are loss averse, the risk of retroactive losses may encourage consumers inertia, or their tendency to wait. For example, a loss can happen when valuation is realized to be lower in the future compared to the level of valuation dur-ing the purchase period. If the consumer values the product less in the future because of unexpected reasons, she will suffer a loss, and therefore she might prefer to wait and to purchase the product when her valuation would be more clear to her. This explanation of the consumers’ loss aversion can be related also to the anticipation of regret in the model of Nasiry and Popescu (2009). Moreover, it is also based on the advance selling methods, when consumers’ valuation is uncertain.
Probability weighting is associated with loss aversion and implies that people tend to give higher probabilities to low-chance situations, or to over-estimate occurrence of low probability events. Su (2009) argues that even when there is a small chance of realizing low valuations, consumers may assume the odds as much higher, an assumption which results from the probability weighting bias. In contrast with the assumption of rationality, due to this bias the consumer will prefer to wait with the purchase.
In order to be able to assume higher probabilities than actually exist, those probabilities should be unknown to the consumers. This may be re-lated to ambiguity aversion, a situation first studied by Ellsberg (1961). Ambiguous probability is basically an unknown probability. This is a sit-uation of uncertainty which involve missing information compared with a risk situation, where the probabilities are known and one can evaluate and calculate the risk of her decisions. The decisions of people in a situation of uncertainty and ambiguity, may reveal the probabilities that they assigned to each of the decision options (Ellsberg, 1961).
Empirical evidences show that people prefer to bet on more certain cases, as they are averse to the ambiguity of other cases (Camerer and Weber, 1992). If we relate this to behavior of consumers in dynamic pricing models, we may deduce that buyers would prefer to purchase a product in the next period, as they will have more information by then.
Kahn and Sarin (1988) use the term ”ambiguity” to define decisions where the odds of an uncertain situation are not precisely known. The
au-thors focus on consumer behavior under the conditions of uncertainty, which is more similar to real life situations, since in real events the exact probabil-ities often cannot be determined, and only a possible range can be assigned. Theoretically, consumers’ decisions should not be different if there is an am-biguity, yet, empirical results show that consumers change their decision pattern when uncertainty is involved (Kahn and Sarin, 1988).
The current study relies on the finding of the studies which were re-viewed in this section, according which behavioral regularities such as loss aversion, probability weighting and ambiguity aversion are the main reasons for delaying purchase behavior. As can be implied from the reviewed stud-ies, researchers do not agree about a clear and distinct bias or act which responsible for such delaying behavior. This behavior is rather derived from strategic concerns or from different behavioral biases, or even from both of them together. The wide ideas and findings show that further research on this subject is needed.
Similarly to some of the studies above, the current study idea is based on posted-prices and advance purchase setting, when consumers buy the product before they are using it, as this situation involves higher valuation uncertainty. Using different levels of valuation uncertainty in the purchase dilemma gives the opportunity to check which one of these factors makes a higher increase in waiting behavior.
It is important to note that most of the papers that describe consumer behavior in dynamic pricing models do so in order to improve the firms’ strategy and to increase their revenues. The consumer perspective is rarely considered as their main focus, and it is the main focus of the current paper. Additionally, most of the studies mentioned above present theoretical models and only two of them presented experimental evidences. These experimental studies did not introduce any other uncertainties to the subjects in addition to prices, which is therefore the contribution of this research.
Providing a controlled experimental evidence for the behavior of con-sumers within dynamic pricing models and increased uncertainty would complement the extensive theoretical models existing in literature so far. Also, it may emphasize other aspects of this behavior. The results from this study are aimed to shed light on the reasons for consumers’ delay be-havior and will help firms to be more accurate in their pricing models and strategies.
3
Experiment
3.1 Method
A lab experiment was used to test the research questions. In the experiment, the participants faced a decision making problem. This problem imitates a
situation in which a consumer wishes to buy a product, in this particular case - a flight ticket for a vacation trip to Berlin in 10 weeks. The price of the ticket is dynamic and may increase or decrease in the following pe-riod. Also, the value which the consumer attach to the product is subject to changes. In dynamic pricing models, prices are subject to change with respect to availability and demand. However, in order to simplify the model and the experiment procedures, in this experiment the prices were decided beforehand and did not depend on subjects’ decisions.
3.1.1 Subjects
A total of 46 subjects participated in the experiment, twenty three subjects for each of the main treatments. Participation was on a voluntary basis and the participants were randomly assigned to treatments. Half of the partic-ipants received an email request for participation and half were recruited in person from the University of Amsterdam library. Twenty five of the subjects were economics students (current students or studied economics in the past) and 21 were students of other departments. There were 33 females and 13 males subjects, and the average age was 23 years old.
3.1.2 Incentives
Subjects could earn money in the experiment depending on their decisions. The experiment use the units ”tokens” as experimental points which were converted to Euros at the end of the session in the exchange rate of 100 tokens = 2 Euros. An average of 2.14 Euros from the possible earnings range of 1.5 to 3 Euros was earned by the subjects in about 15 minutes of participation.
3.1.3 Experimental procedures
The experiment was programmed using the web tool www.thesistools.com. All subjects answered the experiment on a computer. Subjects were asked to read the instructions carefully. each part last for 10 trials, and did not have time constraints. The instructions for each part were presented in the beginning of the part. For the full instructions see appendix A.
After reading the instructions, the subjects were given two example ques-tions in order to check whether they understood the instrucques-tions and the payoff structure correctly (see appendix B for the questions). The questions presented an optional scenario and the participants needed to choose the correct earnings for this particular situation. The subjects received a feed-back for their answers and if they were incorrect, they received a detailed explanation for the correct payoff.
After completing the experiment, participants answered few general de-tails questions. All subjects that were recruited in person to participate in
the library were paid right after they completed the experiment according to their decisions. The rest of the subjects were paid only after few days using an internet banking money transfer.
3.2 Experimental design
Subjects faced two choice options - to buy the flight ticket under the current price, or to wait for the next period’s price. Namely, subjects chose whether to buy in the first or second period. However, for simplicity, the game was played only in one period.
First, the subjects were given an information about the path of the flight ticket’s price so far. This information was presented to them in a graph with the last 10 periods before their time of search (see appendix C for examples). This information helped the subjects to learn about the previous periods’ prices and to form expectations about the price in the next period. Second, the subjects received an information about their private valuation of the product. Their valuation is a variable which is subject to changes, as will be explained in the next section regarding the treatments. The earnings of each subject was calculated as the difference between the valuation and the price paid in the period of purchase.
This experimental designs follows a similar approach which was imple-mented in Chen and Schwartz (2008). Chen and Schwartz concentrate mainly on the role of price patterns and how they shape consumers’ ex-pectations and their booking decisions. Their experimental design includes four treatments: increasing prices, decreasing prices, flat rate, and fluctu-ating prices. Compared with Chen and Schwartz (2008), here I introduce more realistic price patterns in order to observe consumer decision making process that aims to better simulate real world situations.
In the current experiment, the price paths are presented with an upward or downward direction as well, yet with relatively more fluctuations. The patterns are presented only in the last three periods’ prices. Five types of price paths will be taken into consideration: (1) increasing, (2) decreasing, (3) fluctuating from above, (4) fluctuating from below, and (5) flat prices.
3.2.1 Treatments and Hypotheses
This experiment contains three treatments:
(1) Control treatment - price uncertainty - the participants were shown a graph of price history and were given their own current valuation for the product. Their value remained constant through each trial and did not change during this treatment. The subjects were explicitly asked whether they want to buy the product or to wait for the next period. In a situation where the subject decides to wait and the price increased above her value,
her calculated earnings will be zero (hence, she does not have to buy the ticket).
Hypothesis 1 The subjects will learn from the price path, and their de-cisions will be affected from it. When the price path is with up-ward (downup-ward) direction, buying (waiting) behavior will be observed more.
This hypothesis is in line with the findings of Chen and Schwartz (2008) and Yuan and Han (2011). Chen and Schwartz (2008) concluded that the propensity to book is influenced by the price pattens the subjects observed. However, they found that both increasing and decreasing price patterns had the highest booking proportions, with no statistically significant difference between them. In the study of Yuan and Han (2011), the subjects are searching more when the current price is high, but search less if they expect even higher price. Therefore, when they observe increasing pattern, the subjects might expect the next period price to be higher, which will make them buy the flight ticket immediately.
This treatment formulates a single source of uncertainty - the prices. I expect that the subjects will not act as myopic, and will decide to wait even when their valuation is above the price and their earnings are positive. Namely, they will act strategically in order to increase their expected utility. This treatment will be used as my control to compared with the next two treatments. It mainly involves the formation of price expectation of the sub-jects, and aimed to observe the subject decisions based on these expectation. (2) Loss Aversion treatment (LA) - double uncertainty, prices and val-uation may change - the participants were shown a graph of price history and were given their own current valuation, and a probability for a change in their valuations in the next period, over a known range. As in the first treatment, the subjects were asked whether they want to buy the product or to wait for the next period. Earnings in this treatment were calculated according to the realized valuation in the next period, so the subjects may suffer a loss if their valuation had decreased while prices have increased in the next period. However, if they decided to wait, and the outcome in the following period reflects a loss (as described above), then they are not obligated to buy the product, so they will earn zero tokens.
Hypothesis 2 When there is a probability for a change in valuation and a risk of loss, subjects will decide to wait more compared to the control treatment.
The higher the probability for a loss, the higher the tendency to wait for valuation realization as in the study of Su (2009). In the control treatment
the probability for a loss is zero, as the value is constant. Thus, in the loss aversion treatment waiting behavior is expected to be positively correlated with possible change in valuation.
(3) Ambiguity Aversion treatment (AA) - uncertainty in price and val-uation, and the probability for a change in valuation is unknown to the participants - the participants were shown a graph of price history and were given their own current valuation for this period. The computer randomly assigned a probability for a change in their valuation in the next period, which was unknown to the participants. Although the probability is un-known, the expected utility in this treatment is still the same as in the LA treatment, following the ambiguous strategy which was presented in Sarin and Weber (1993). In this treatment, the subjects were expected to form their own beliefs about the probability for a change in the valuation, in addition to their beliefs regarding the price. As was evident in past liter-ature, the subjects might overestimate the probabilities. Because of that, their behavior is assumed to be different from the LA treatment, in which the probabilities are known, and from the control treatment, in which the probability is zero and values are constant over all periods.
Hypothesis 3 In this treatment subjects will wait more than in the control treatment, due to the increase in uncertainty.
Hypothesis 4 Comparing the two treatments (LA and AA), I expect that subjects will choose to wait more in the ambiguity aversion treatment compared to the loss aversion treatment as they will have higher level of uncertainty which is expected to result in waiting behavior.
These two main treatments examine the implications of different degrees in uncertainty on delaying behavior. Each one of the treatments adds an-other layer of uncertainty to the consumers’ problem. In the first treatment, which I define as the control, the subjects are facing only one source of uncer-tainty, prices, since in dynamic pricing models prices are subject to changes constantly. In the second treatment the subjects face more uncertainty in their own valuation of the product. This allows including an important component of dynamic pricing models, which often involve purchasing be-fore consuming, when real valuation is not yet realized by the consumers. The last treatment makes the situation even more similar to the real events, as it involves ambiguity in the probability of change in valuation. When a consumer buys a product before using it, she might later experience some changes in her valuation. However, the probability for a change is unknown to her, and this treatment attempts to simulate this uncertainty.
Including these different levels of uncertainty in the experiment is useful in the sense that it can teach us about the dominance of these biases, the loss aversion and the ambiguity aversion, in consumer decisions in dynamic
pricing situations. I expect that each treatment will reveal higher propensity of waiting behavior depending on the level of uncertainty introduced.
Moreover, consumers’ decisions may also be affected by learning from the price history. They can form beliefs about the probability that the price will increase or decrease in the next period, hence we must consider it while estimating their decision making process.
Each subject will be exposed to the control treatment and only to one of the LA or AA treatments. Hence, a within-subject analysis will be con-ducted between the main treatments and the control treatment, as well as between-subject analysis for comparison between the main treatments. For a better comparison between the treatments, the graphs that were presented to the participants in the LA and the AA treatments are the same, only the information about the probability is different. Figure 1 below summarize the experimental design and the comparable relations between the treatments.
Figure 1: The experimental treatments and the comparable relations between them
4
The theoretical decision rule
To explain the decision process of the subjects in the experiment I will first present a simple decision framework, according to a rational decision making theory.
Let U define the utility from purchasing the flight ticket in the current period. Similarly, the utility from purchasing the flight ticket in the next period (after the subject decided to wait) will be defined as U0.
In general, U and U0 are equal to:
U = (
V (α) − P0, if V (α) ≥ P0 η(V (α) − P0), if V (α) < P0
and
U0= V (α) − fP1 where fP1 = fi(P0) (2)
Where V (α) is the subject’s expected value of the flight ticket, which depends also on an uncertainty parameter, α. P0 and fP1 are the prices in the current period and in the next period, respectively. The subjects do not know the value of P1 when they are making their decision, therefore, U0 depends on fi(P0), their individual beliefs about P1 after having the information regarding P0. U is identical over all subjects since they all have the same values and prices, however, U0 may change between subjects according to their beliefs regarding P1.
As there is a chance to lose if the subject buys the product and her valuation appeared to decrease below P0, I added η to the utility function which is the loss aversion coefficient (η > 1). This specification also follows the decision model described in Su (2009).
Three possible situations need to be considered. i V (α) < P0 and V (α) < fP1
If V (α) is below P0, then the rational consumer will not buy the product and will certainly wait, since this case involves negative earnings. If the price in the next period, fP1, is also higher than V (α), the subject will not buy the product in the next period as well and will have zero earnings. ii V (α) < P0 and V (α) > fP1
In the second case, if V (α) is below P0, the consumer should wait for the next period for sure, but if fP1 is below V (α) the consumer will buy the ticket in the next period, as she has a positive utility.
iii V (α) > P0
The third situation is the most relevant for the discussion of this study. If V (α) is higher than P0 the consumer is facing the question whether to buy the flight ticket now or to wait.
Thus, for the situation of V (α) > P0, the decision rule of the subject is as follows:
wait = (
1, if U0 > U
0, otherwise (3)
we can replace the terms in the decision rule with the definitions from (1) and (2). Then we will get:
U0 > U → V (α) − fi(P0) > V (α) − P0 → P0 > fi(P0) (4) or, in general form which would reflect also a situation which fP1 < V (α) < P0 applies:
V (α) − fi(P0) > ηV (α) − ηP0 P0 >
fi(P0) + (η − 1)V (α)
η = F (η) (5)
In the control, η = 1 and V (α) is always higher than P0, so equation (5) is reduced to equation (4). Hence, when the price of the next period is believed to be smaller than the price in the current period the subject should decide to wait.
According to the decision rule, when the subjects have the option to lose, they will wait if equation (5) is satisfied. As the hypotheses of the study suggest, the subjects are expected to make more wait decisions in the treatments compared to the control due to the increase in uncertainty and loss aversion.
The expression in (5) depends on the η level of the subjects, or how loss averse they are. If we derive this expression with respect to η we will get the following:
∂F (η)
∂η =
fi(P0) − V (α)
η2 (6)
If fi(P0) > V (α) (subjects believe that future prices will be higher than their valuation) then the subjects will not buy the ticket. However, when fi(P0) < V (α), the subjects have incentives to buy the ticket. In this situ-ation, ∂F (η)∂η < 0, which means that if η is increasing, the expression F (η) is decreasing. Hence, according to the decision rule, the higher the level of which a subject is loss-averse, the more likely she is to display waiting behavior. This is because the equation in (5) can be satisfied under lower levels of the initial price P0.
The expected value of the subject is taking a general form which changes in each treatment. However, the decision should still be based only on the subjects’ beliefs regarding the prices, as we can conclude from (4) and (5).
The expected value, V (α), is calculated according to the following: V (α) = (1 − α)VCurrent+ α[
VL+ VH
2 ], VNext∈ [VL, VH] (7)
Where VCurrent is the current value, VL is the lower bound of the value range and, VH is the upper bound of the value range. Here, α represents the probability that the subject’s value will change in the next period over a given range, or the uncertainty factor in the value.
In the experiment, I used a constant value range. The parameters a and b are non-negative constants which create the value range according to the following:
VL= Vcurrent− a and VH = VCurrent+ b (a, b > 0, b < a) (8) Thus
V (α) = VCurrent+ α[ b − a
2 ] (9)
In the control treatment, α is equal to zero in all trials, since the value is constant. Thus, the subjects have complete information about their value in the next period and need to consider only the change in prices while making their decision. In the loss aversion treatment, the probability is given to the subjects, so α is known and varies between 0 − 100%. In the ambiguity aversion treatment α is unknown, but it still exists and depends on the subjects’ beliefs.
Even though in the LA and in the AA treatments there is a probability that the subject’s value will change, the expected value is always set higher then P0, meaning, completely similar to the control part. Because of that, the decision rule is the same for all treatments, as the expected value is canceled out from equation (4).
Hence, if the subjects would decide to wait more in the loss aversion or in the ambiguity aversion treatments (compared to the control treatment), we can conclude that the increased uncertainty also plays a role in the decision rule of the experimental problem, and that it does not only depend on the price as rational decision theory would predict.
5
Results
In order to test the hypotheses of this experiment, nonparametric tests were used to evaluate the differences in subjects’ behavior between the treatments and logit regressions were used to estimate the propensity to wait of the subjects.
5.1 Nonparametric tests for comparison between treatments I begin with estimation of the differences in decisions between the treat-ments as hypotheses 2, 3 and 4 suggest, using the Wilcoxon nonparametric test. For the within-subject analysis I used Wilcoxon Signed-Rank test, and Wilcoxon Mann-Whitney test was used for the between-subject analy-sis. This nonparametric test holds the null hypothesis that two samples are from the same identical distribution against the alternative hypothesis that the two samples differ with respect to their median.
Table 1 provides a raw data summary of wait decisions of each subject in each of the treatments. Table 2 provides a summary of the differences
between the treatment parts to the control part and between the two treat-ments. Column 4 of this table provides the results for the within-subject analysis and row 4 provides the results for the between-subject analysis.
In the loss aversion treatment, the average of wait decisions is 5.09 (out of 10 optional decisions) compared to an average of 5.22 wait decisions of the same subject’s group in the control part. The difference between the averages (-0.13) is found to be not significant at any significance level using a one-tailed Wilcoxon Signed-Rank nonparametric test (p = 0.48). Following this result, we can reject hypothesis 2 and conclude that subjects did not increase their waiting behavior due to the risk of losing in the loss aversion treatment.
Table 1: Summary data: Number of ”wait” decisions, by subject
Loss aversion treatment Ambiguity aversion treatment Participant number Control graphs Treatment graphs Participant number Control graphs Treatment graphs 1 4 4 24 8 5 2 6 3 25 7 6 3 4 5 26 4 6 4 2 3 27 6 4 5 6 5 28 5 5 6 6 4 29 5 8 7 5 5 30 2 5 8 4 2 31 5 4 9 7 7 32 5 5 10 5 5 33 9 8 11 7 8 34 3 2 12 6 10 35 3 5 13 5 9 36 3 6 14 7 4 37 8 9 15 5 5 38 5 7 16 7 6 39 5 6 17 4 3 40 4 9 18 5 6 41 5 6 19 4 3 42 4 4 20 5 4 43 4 3 21 3 3 44 4 8 22 8 7 45 4 10 23 5 6 46 6 5 Average 5.22 5.09 Average 4.96 5.91
Notes. Cell observations are the sum of the number of waiting decisions in each of the part graphs out of 10 trials of each part. The last raw gives the average of waiting decisions for all subjects in the same treatment.
The average of wait decisions in the ambiguity aversion treatment found to be 5.91 compared to 4.96 wait decisions of the same subject’s group in the control part. This nearly 20 percent increase in waiting behavior is
significantly different from the behavior in the control part at the p < 0.1 level using a one-tailed Wilcoxon Signed-Rank test, which has the alterna-tive hypothesis of the ambiguity aversion treatment having on average more waiting decisions (p = 0.097). The results of this test allow us to not reject hypothesis 3, as we can conclude that subjects decided to wait more in this treatment compared to the control treatment.
Table 2: Summary statistics - Average of ”wait” decisions by treatment and the differences between them
Average Control Average treatment Difference
AA 4.96 5.91 0.96*
(0.36) (0.42)
LA 5.22 5.09 -0.13
(0.30) (0.43) Difference -0.26 0.83
Notes. The cell observations give the mean and standard errors of the wait decisions for each participant over the 10 trials of each treatment. Column 4 and row 4 provides the differences between the average of wait decisions. Column 4 provides the differences between the treatments to the control, and row 4 gives the differences between the treatments. These values also indicated whether the difference is significant at the p < 0.05 (**) or p < 0.1 (*) using a nonparametric Wilcoxon test.
Conducting a between-subject analysis, I compare the behavior of the two groups of subjects. Using Wilcoxon Mann-Whitney nonparametric test, we can conclude also that there was no significant difference between the be-havior in the control part of the two subjects groups (see row 4 of table 2). Since the subjects were randomly assigned for treatments, and the control part is the same for both of the groups, there is no reason to assume differ-ence between the decisions in this part of the experiment. Therefore, this result gives a good evidence for the random procedure of this experiment.
However, according to the results, no significant difference was found between the two treatments, the loss aversion and the ambiguity aversion, disproving hypothesis 4. Though not significant, the difference between the waiting decision averages, 0.83, is compatible with the correct direction of the hypothesis, which assumed that subjects will be more likely to wait in the ambiguity aversion treatment than in the loss aversion treatment.
Figures 2 and 3 present the percentage of waiting decisions of each treat-ment group per graph of each part. We can see that in some graphs subjects generally prefer to wait with their purchase and in other graphs the pref-erence to buy was found to be much stronger. The diffpref-erences between the decisions of the two treatment groups for each graph are not significant using the Wilcoxon nonparametric test. Significant difference is observed only in graph number 5 of the control trials (LA > AA, z = −2.653, p = 0.008), and
in graph number 8 of the treatment trials (AA > LA, Z = 2.393, p = 0.017). Figure 2: Percentage of ”wait” decisions in the control part
by trial and treatment group
Figure 3: Percentage of ”wait” decisions in the treatment part by trial and treatment group
5.2 Logit model estimations - The probability to wait
In addition to assigning treatment definitions to each trial, the trials in the experiment also have different characteristics: value, price and pattern. As can be seen from figures 2 and 3, differences in trial characteristics may have influenced the choices of the participants. These differences between trials allow estimation using the logit model. The logit model can be used as an
efficient tool in order to evaluate which of the parameters had an effect on the propensity to wait of the subjects (Train, 2009). The subjects in the experiment faced a binary choice option, which forms the framework for a logit model estimation.
Also, this estimation allows me to test the validity of hypothesis 1, as it will provide an answer to the question whether the subjects were able to learn from the price paths. Actually, hypothesis 1 is the basics of the subjects’ decision in dynamic pricing models, before they were manipulated with the valuation uncertainty in the experiment.
I start with analyzing the decisions of the subjects as they were reflected from the control graphs. In those graphs the only uncertainty the subjects are facing is the one of the price which may decrease or increase in the next period. The subject’s value is constant over the decision period and in the next one.
The estimated model is formulated as follows:
Pi(wait = 1) = F (β0+ β1∗ X1+ β2∗ X2+ ... + βn∗ Xn+ i) (10) where:
F (z) = e
z 1 + ez
This model holds the null hypothesis that all β’s are equal to zero. Ac-cording to the hypotheses of this study and the literature reviewed before, I am expecting that the direction of the coefficients will be as follows.
First, I will estimate the logit model using the absolute values. The coefficient of the ”current price” variable is expected to be positive, since when the price is higher there is a better reason for the subject to wait. Compared to that, the coefficient of the ”value” variable, is expected to be negative. If the value of the subject is high, it mean that she has more earnings, keeping all other variables as constants, therefore she might decide to buy the product immediately and not to wait for the next period. The results of this specification are presented in column (1) of table 3. The odds-ratios presented in the table were calculated according to eβ. In all reported model estimations, standard errors were clustered based on each subject.
As was expected, the coefficients of the price and the value are signifi-cantly positive and negative respectively. We may infer that the higher the price, the more likely it is that a subject would wait for the next period to buy the flight ticket. If the price would increase in one unit, the odds of deciding to wait would increase by a factor of 1.26 (26% increase). On the other hand, if the value would increase in one unit, the odds that the subject would decide to wait would increase only by the factor of 0.8, or namely, decrease by about 20%.
The graphs pictures which were showed to the participants in the exper-iment had an end-pattern, whether the prices are in an increasing direction,
Table 3: Logit regression estimation - Control graphs, all subjects
(1) (2)
VARIABLES choice Odds-Ratios choice Odds-Ratios
price 0.229*** 1.257 0.245*** 1.278 (0.0468) (0.0470) value -0.221*** 0.802 -0.238*** 0.788 (0.0454) (0.0456) pattern 2 0.312 1.366 (0.353) pattern 3 0.00459 1.005 (0.338) pattern 4 0.474 1.606 (0.294) pattern 5 -0.148 0.862 (0.305) Observations 460 460
Notes. Robust standard errors in parentheses. Variances are adjusted using a cluster on subject. Odds-Ratios were calculated according to eβ. *** p < 0.01,
** p < 0.05, * p < 0.1
decreasing, flat and two types of fluctuation. The pattern of the price path might have also an effect on the decision of the subject, following Chen and Schwartz (2008). Column (2) of table 3 presents the results of the logit model for the control part with the pattern variable. However, none of the coefficients of the patterns categories compared to the first pattern is significant. Therefore, I will not provide interpretation for these results.
Another option to test the effect of the price patterns on the propensity to wait is replacing the categorical ”pattern” variable with a variable which represent the price of the period before the current price, and the price of the period before that. This specification is more accurate addition of the pattern variable, considering that the pattern was only appeared in the last three periods’ prices in each graph. A similar estimation was used in the study of Yuan and Han (2011). The results of this specification are presented in table 4.
According to the results in table 4, the price of the period before the decision period, ”price−1”, had a positive and significant effect at the 10% level on the subjects’ probability to wait. The price of the period before that, ”price−2”, had a significant negative effect. Namely, if the price of the period before the decision period or the price before that will increase in one unit, the odds that the subject would decide to wait would change only by the factor of 1.03 (3% increase) and 0.976 (2.2% decrease) ,respectively.
Table 4: Logit regression with the prices of past periods -Control graphs, all subjects
(1)
VARIABLES choice SD Odds-Ratios
price 0.263*** (0.053) 1.301
value -0.259*** (0.049) 0.772
price−1 0.029* (0.018) 1.030
price−2 -0.024* (0.014) 0.976
Observations 460
Notes. Robust standard errors in parentheses. Variances are adjusted using a cluster on subject. Odds-Ratios were calculated according to eβ. *** p < 0.01, ** p < 0.05, * p < 0.1
The addition of the previous price variables did not change the effect of the current price and the value variable significantly.
Next, I test whether the difference between the value and the price has an effect on the subject decision to wait. This difference defines the utility from buying now, so it might be considered as a more relevant variable to evaluate the propensity to wait. The results of this estimation, which are presented in table 5, show that if the difference between the subject’s value and the current price will increase in one unit, the odds that the subject would decide to wait would change by the factor of 0.97 (decrease in 3%), in a 10% significance level. After the prices of the previous periods were added, the model had a better explanatory power (p < 0.00). As can be seen from column (2), in this specification, the odds that the subject would decide to wait would change by the factor of 0.77, with p < 0.01, or decrease in about 23%. Adding the differences between the prices, a variable which might be similar to the ”pattern” variable, generates a non-significant model. Therefore, its results, which are presented in column (3), are invalid for interpretation.
From the above estimations we can conclude that the subjects are able to learn from the prices and that their decisions are affected by them, and by the relations between their own valuation and the current price. There-fore, we can not reject hypothesis 1, which means that subjects are able to learn and to form their expectations based on the information they ob-served. However, the hypothesis also suggests that if the prices are increasing (decreasing) the subjects would tend to buy (wait) more. Testing this
hy-Table 5: Logit regression estimation of the difference between the value and the price
-Control graphs, all subjects
(1) (2) (3)
VARIABLES choice Odds-Ratios choice Odds-Ratios choice Odds-Ratios Value - Price -0.027* 0.97 -0.26*** 0.77 -0.025* 0.975 (0.015) (0.05) (0.015) price−1 0.032** 1.032 (0.014) price−2 -0.023* 0.977 (0.014) price-price−1 -0.006 0.994 (0.014) price−1-price−2 -0.012 0.988 (0.013) Observations 460 460 460
Notes. Robust standard errors in parentheses. Variances are adjusted using a cluster on subject. Odds-Ratios were calculated according to eβ. *** p < 0.01, ** p < 0.05, *
p < 0.1
pothesis using a logit model estimation resulted in a non significant model. Accordingly, we cannot determine whether the patterns, and not the abso-lute values of the prices, had a significant effect on the decisions. Therefore, the hypothesis was only partially accepted.
One issue to be considered while interpreting the results of this estima-tion is that it does not take into account the subjects’ beliefs regarding the price of the next period. If they decide to wait, their utility and earnings would be determined by the last period’s price. Hence, the decision is also based on this unobserved variable. The reason for estimating the model without the beliefs variable is to give an impression of how the known vari-ables (which the subjects observe) may affect their behavior. The results of the estimation reveal the subjects’ implied beliefs about the next period price.
6
Discussion
In this section I will discuss potential explanations for the experimental results. The results presented in section 5 show that only the ambiguity aversion treatment had a significant effect compared to the control part. Therefore, we can not reject the hypothesis according which the subjects
will wait more in this treatment compared to the control part. The in-crease in uncertainty and the ambiguity stimulate the waiting behavior of the subjects.
The uncertainty in this treatment was designed to be higher than in the loss aversion treatment and in the control part, as the subjects did not know what is the probability for a change in their valuation of the flight ticket. Not knowing the exact probability increased the uncertainty, yet according the decision rule presented before, it does not suppose to make a difference in the decision process of the subject.
Few possible explanations can be provided for why no increase in waiting behavior in the loss aversion treatment was found. First, it is possible that while solving the problems of the loss aversion treatment, when the proba-bility was known, the subjects felt like they have sufficient information in order to calculate the chances and the expected value. Therefore, they might have been trying to use the new information and to become more ”strate-gic” while facing the loss-aversion treatment situations. If they indeed were able to determine the expected value, they might behave according to the rational theory. This explains the low average of waiting decisions in this treatment compared to the control part and the ambiguity treatment.
Moreover, some experimental design issues may also be considered as possible explanation for the results of the loss aversion treatment. If we return to the theoretical decision rule model, in this experiment the option of V (α) < P0 was not given at all, in order to avoid situation where the choice of waiting was clearly superior. Even though all values, prices and probabilities were randomly produced by the computer, calculation of V (α) always gave a higher expected value compared to the price of the current period (except of one graph in the treatment part). Such condition implies that the subjects did not have the option to lose, so loss aversion did not take a significant role in the decision process in this treatment.
However, in the ambiguity treatment, the subjects could not calculate the expected values, as they did not have the complete information about the probability. Calculation of the expected value was based only upon their beliefs. It is possible to think that the subjects formed beliefs about the probabilities, which are higher than the actual probabilities, as they significantly decided to wait more. If they did so, then the expected values they had might be lower than P0, a situation that involves the loss aversion utility. Therefore, the loss aversion bias takes a role in this treatment, and the effect of it is shown in the significant result. Consequently, we can conclude that the subjects were able to understand that waiting for valuation realization might be the optimal decision, since it implies that they will never lose their earnings.
Overall, the increase in the uncertainty did not change the subjects’ behavior dramatically. Another reason for it could be that the uncertainty is an integral part of purchasing a product with dynamic pricing, since
it appears in the constant price changes. There is a possibility that the constant changes in prices are the most powerful source of uncertainty in this situation and because of that, the additional uncertainty in the valuation did not make much different, only when it was extremely high as in the AA treatment.
For this reason, I used the logit model to estimate the effect of the current price level on the subjects decisions. As seen in the results, the probability that the subject will wait is higher when the level of the current price increases. For example, it is easy to imagine a situation in which a price is too high at the moment. Waiting for few days might not make a lot of difference as the price is already too high, so it is preferable to wait with hope that it might decrease. In addition, a more precise decision will be made according to the difference between the valuation and the price as the results from table 5 give us. If the current utility is already high, one should buy the ticket immediately. Since waiting also involves some risks, sometimes buying can be the optimal decision.
The absence of a statistically significant effect of the price pattern can be explained by the possibility that the price patterns that were presented to the subjects were not clear enough. The fluctuations in prices may have seemed to be too volatile, which may have made it harder for the subjects to understand the path of prices from it, and encourage them to base their decisions according to the current price. Another option is that the subjects did not observe it as a meaningful variable. Third, if we look at the price patterns in the real world, just recently search platforms started to present the path of prices, so until then consumers were only familiar with one price - the price in the current period.
7
Conclusions
The results presented in this research demonstrate that uncertainty levels have an effect on consumer behavior in dynamic pricing situation, yet only in situations when the uncertainty is very high. In fact, ambiguity in the probability for a change in valuation is a situation that commonly occurs in real life. In real life the consumer is never aware of the exact probability and choices fully depend on each consumer’s beliefs. Hence, a high uncertainty level is more similar to a ”real” flight booking situation, and this controlled study potentially provides us more knowledge about the decision making process of the consumers under uncertainty when facing a dynamic pricing products. Furthermore, it is reasonable to believe that the effect of the uncertainty would be even more significant in a real situation.
Although the experiment tried to provide the subjects with a real world dilemma, it cannot be ignored that the behavior in the experiment is likely to be different from the behavior outside of the lab. Compared to real actors,
the subjects in the experiment may present a more risk-loving behavior in order to increase their earnings, while in the same time they have ”nothing to lose”, despite the fact that the experiment was incentivised. Consumers of flight tickets in the real world eventually wish to book the flight ticket, and will not be satisfied with not having a seat on this flight, or as the experiment presented it with ”zero earnings”. Therefore, they might be more risk averse than the subjects in the experiment.
In order to give more attention to the risk attitudes of the subjects, it might be interesting for future research to measure the risk attitude of each of the subjects before conducting the experiment and to try and classify them into groups of risk averse or risk loving. By that, we will be able to better observe whether the effect is different for each type of the consumers. These differences may be especially relevant in the particular case of purchasing a flight ticket, as usually flights are very expensive and include a high one-time expense on the flight, but after that it includes additional travel expenses such as accommodation etc. As a result, risk behavior may take a more essential role in the decision making process.
Moreover, in the current experiment I decided to focus only on few se-lected elements involved in the decision process of dynamic pricing products. This was done mainly for simplicity reasons and in order to be able to dis-tinguish between the elements that affect the decision using a controlled environment. However, examining other factors may give a wider percep-tion about the process. For example, the availability factor might increase the buying behavior of consumers because of the fear from not being able to buy the product later, as it will not be on stock any more. Also, the pur-pose and context of purchasing the product might also change the behavior. For instance, in flights, leisure consumption might induce different behavior patterns from business consumption. Another interesting comparison might be experienced consumer behavior compared to unexperienced. An experi-enced consumer might be more inclined to wait and not to buy the ticket immediately, maybe because his last flight booking was a ”failure” and he was able to learn from his experience. Moreover, in this experiment I used only a single time of arrival to the market for all subjects, 10 weeks before the departure. However, it would be interesting to see whether consumers change their delaying behavior with respect to their time of arrival to the market.
Future research might also benefit from conducting a field experiment, which will use real consumers as subjects and therefore will allow greater generalization of the results. The more realistic settings of such field exper-iment may provide new insights to the decision process, which will include all variables that are involve in it.
In summary, the ”right” strategy to buy a product with a dynamic prices is not clear. Usually waiting with the purchase will decrease the uncertainty involve in the process. However, delaying purchase may also result in an
un-wanted outcome, such as increase in prices or low availability. This complex decision process involve a lot of ”what if” questions and different factors are needed to take into consideration. As a consequence, the over-uncertainty associated with those products’ purchase might even cause the consumer to give up trying to make a strategic decision and to choose the purchase time randomly rather than according to some rule of thumb.
The current study gives an experimental evidence for this decision pro-cess, which was so far mainly discussed in theoretical literature. Further investigation of the behavior of consumers, as was suggested above, might tell us more about the decision making process of this complex situation.
Understanding this process more thoroughly is a mutual interest of both consumers and firms. Recently, firms began to understand that consumers might be afraid to commit, and that their struggle with the question of when to buy leads them to attempt to time their purchases. As it affects the firms’ strategy as well, they are trying to provide different decision tools to the consumers. The main reasons for it are that they want to answer this demand efficiently and to make higher revenues, but it can also contribute to the consumers by making their decision process a bit easier and information-based. Tools and options such as flexible cancellations policies, benefits from advance booking, saving rates opportunities and more are only partial exam-ples for implementation of this understanding regarding consumer behavior in dynamic pricing situations. Of course, all of these tools involve additional costs to the consumer, but for many consumers paying for having the option to decrease the uncertainty is worthwhile.
References
Anderson, C., Wilson, J., 2003. Wait or buy? The strategic consumer: Pric-ing and profit implications. Journal of the Operational Research Society 54 (3), 299–306.
Bitran, G., Caldentey, R., 2003. An overview of pricing models for revenue management. Manufacturing & Service Operations Management 5 (3), 203–229.
Camerer, C., Weber, M., Oct. 1992. Recent developments in modeling prefer-ences: Uncertainty and ambiguity. Journal of Risk and Uncertainty 5 (4), 325–370.
Chen, C., Schwartz, Z., Apr. 2008. Room Rate Patterns and Customers’ Propensity To Book a Hotel Room. Journal of Hospitality & Tourism Research 32 (3), 287–306.
Ellsberg, D., 1961. Risk, ambiguity, and the Savage axioms. The Quarterly Journal of Economics 75 (4), 643–669.
Elmaghraby, W., Keskinocak, P., 2003. Dynamic pricing in the presence of inventory considerations: Research overview, current practices, and future directions. Management Science 49 (10), 1287–1309.
Gallego, G., Ryzin, G. V., 1994. Optimal dynamic pricing of inventories with stochastic demand over finite horizons. Management science 40 (8), 999–1020.
Kahn, B., Sarin, R., 1988. Modeling ambiguity in decisions under uncer-tainty. Journal of Consumer Research 15 (2), 265–272.
Levin, Y., McGill, J., Nediak, M., Jan. 2009. Dynamic Pricing in the Pres-ence of Strategic Consumers and Oligopolistic Competition. Management Science 55 (1), 32–46.
Li, J., Granados, N., Netessine, S., 2011. Are consumers strategic? Struc-tural estimation from the air-travel industry. Working Paper.
Liu, Y., Cooper, W., 2014. Optimal Dynamic Pricing with Patient Cus-tomers. Working Paper, 1–49.
Nasiry, J., Popescu, I., 2009. The price of consumer regret. Working Paper, 1–45.
Nasiry, J., Popescu, I., 2011. Dynamic pricing with loss-averse consumers and peak-end anchoring. Operations research 59 (6), 1361–1368.
Ovchinnikov, A., Milner, J., 2012. Revenue Management with End of Pe-riod Discounts in the Presence of Customer Learning. Production and operations management 21 (1), 69–84.
Popescu, I., Wu, Y., 2007. Dynamic pricing strategies with reference effects. Operations Research 55 (3), 413–429.
Sarin, R., Weber, M., 1993. Effects of ambiguity in market experiments. Management science 39 (5), 602–615.
Shen, Z., Su, X., 2007. Customer behavior modeling in revenue management and auctions: A review and new research opportunities. Production and operations management 16 (6), 713–728.
Su, X., 2009. A model of consumer inertia with applications to dynamic pricing. Production and Operations Management 18 (4), 365–380. Talluri, K., Ryzin, G., 2006. The theory and practice of revenue
manage-ment. Kluwer Academic Publishers, Boston/Dordrecht/London.
Train, K. E., 2009. Discrete choice methods with simulation. Cambridge university press.
Tversky, a., Kahneman, D., Sep. 1974. Judgment under Uncertainty: Heuris-tics and Biases. Science (New York, N.Y.) 185 (4157), 1124–31.
Tversky, A., Kahneman, D., 1991. Loss aversion in riskless choice: A reference-dependent model. The Quarterly Journal of Economics 106 (4), 1039–1061.
Yuan, H., Han, S., 2011. The effects of consumers’ price expectations on sellers’ dynamic pricing strategies. Journal of Marketing Research XLVIII (February), 48–61.
A
Experiment instructions
Welcome!Today you are about to participate in a decision making experiment. The details of the experiment are discussed in the next pages. This experiment contains two parts. The instructions for each of the parts will be presented in the beginning of the part. Please read the instructions carefully at your own pace.
In this experiment, you have the option to earn money depending on your choices. The units of exchange in all the trades will be referred to simply as tokens. At the end of the experiment, your cumulative earnings in tokens will be converted into money at the exchange rate of 100 tokens = 2.0 Euros, and which will be paid to you after.
After completing all parts of the experiment you will be asked to disclose some general information details. However, during the course of this exper-iment you will remain anonymous.
Please press START to proceed to the instructions of part one. PART ONE:
You are playing the role of the buyer.
You would like to go on a vacation trip to Berlin in 10 weeks, and therefore you wish to book a flight ticket.
The computer plays the role of the airline seller. The seller posts a price for the flight ticket and promises to supply the product to all buyers who are willing to pay the posted price. The buyers cannot negotiate the price with the seller.
Each week is considered as a time period in this experiment.
You will be presented a graph with the previous prices of the same ticket, and the current price. Moreover, you will receive a personal value for this flight ticket. Your assigned personal valuation reflects the value that you personally attach to the product. Your valuation of the ticket will remain constant in the current period and the next one.
You have two options:
1. BUY - book the flight ticket now.
2. WAIT - wait for the next period and buy it then.
After you have made your decision you will be redirect to a screen that will show you the result of your choice.
