Unpacking the Effects of Weather on Willingness to Pay:
Evidence from eBay
Moshe A. Barach
McDonough School of Business, Georgetown University *
November 7, 2016
Abstract
Using weather as a proxy for mood, extant research has studied the effects of mood on economic
outcomes. However, this literature has generally focused on assessing the direction of the effect
of weather induced mood on economic behavior not the mechanism through which mood affects
behavior. Using a series of empirical tests, I isolate and identify the mechanism(s) through which
weather might influence willingness to pay on eBay.com. I find evidence that rain influences mood,
which alters the perceived value of the item, increasing an individual’s willingness to pay across auctions for an identical item by about 1.5%.
Behavioral Economics, Judgement and Decision Making
* Author contact information, datasets and code are currently or will be available at http://www.moshebarach.com. I would
like to thank Ming Leung for his helpful comments as I was preparing this document.
1
1 Introduction
Both economics and psychology have long linked mood or specific emotions1 to economic outcomes
(Loewenstein and Lerner, 2003; Forgas, 1995). Weather is commonly used as a proxy for mood, as there
is substantial psychological evidence that weather affects mood (Kööts et al., 2011; Denissen et al., 2008;
Hannak et al., 2012). Seasonal affective disorder, more commonly known as the winter blues, is a welldocumented disorder. To some extent, nearly everyone experiences some differences in mood between
warm, sunny, clear days and cold, cloudy, rainy days (Nelson, 2005). Thus, understanding the effects
of weather induced mood variation on economic outcomes can provide great insight into seemingly
irrational behaviors of both buyers and sellers.
Extant research using weather as an exogenous mood shock has focused on assessing the direction
of the effect of weather induced mood on economic behavior not the mechanism through which mood
affects behavior (Coleman and Schaefer, 1990; Markham and Markham, 2005; Lamare, 2013; Busse et al.,
2015). Many studies which relate weather to economic outcomes fail to adequately test if the results are
driven by weather-induced mood or direct weather-related effects. Only a small amount of experimental research has been able to isolate which mechanism(s) might explain the relationship between mood
and economic behaviors such as valuation in the real world. Additionally, this literature has focused
mostly on investment goods, and highlighted the link between weather, mood, and risk-taking behavior in financial decisions (Bassi et al., 2013). Understanding the mechanism(s) through which weather
influences economic behavior is necessary to understand and maybe correct for weather related bias in
decision processing in organizations (Bachkirov, 2015; Brief and Weiss, 2002; Tsai et al., 2007; Seo et al.,
2004).
The literature in psychology has highlighted multiple mechanisms though which mood can affect
economic behavior. (1) Mood has been shown to affect cognition, which can affect economic behavior
(Lazarus, 1991; Smith and Ellsworth, 1985). For example, positive moods like happiness could promote
simpler cognitive processing (Loewenstein, Weber, et al., 2001; Schwarz et al., 1991) or conversely, posi1 Mood and emotion are not the same. According to Tan et al. (2009), mood is "the appropriate designation for affective states
that are about nothing specific or about everything-about the world in general." While emotions which are generally much
shorter in length than moods are a generally a reaction to a specific event and are thus generally about something specific. As
I seek to understand the effects of weather on economic outcomes, I focus on mood as a broad affect state (e.g. good and bad
mood).
2
tive affect could also promote more efficient and adaptive cognitive processing (Isen, Rosenzweig, et al.,
1991; Wegener et al., 1995). (2) Mood has been shown to affect the assessment of risk which directly
alters valuation. For example, Johnson and Tversky (1983) and Kuhnen and Knutson (2011) show that
good mood creates overoptimism and a tendency to attach high probabilities to desirable risky events
and low probabilities to undesirable ones. This could lead to an increase in WTP (willingness to pay)
for risky assest. On the other hand, in order to improve negative mood, people might pursue higher risk
options that are more likely to reduce negative mood (Lerner et al., 2004; Raghunathan and Pham, 1999).
(3) Finally, mood can influence an individual’s assessment of monetary value by affecting preferences.
An individual might construct preferences on the spot when asked to reach a particular judgment or decision (Payne et al., 1999; Capra et al., 2010). A positive mood may cause an individual to focus his or
her attention on positive attributes of an item and assign it a higher valuation (Forgas, 1995). Conversely,
consumer behavior could be at some level driven by unconscious needs to alter bad mood or promote
good mood leading to a higher valuation under negative mood (Winslow, 1986; Billig, 1999; Isen and
Geva, 1987).
To isolate and study the effects of weather induced mood on economic behavior such as valuation or
willingness to pay, I analyze the bidding behavior of individuals on eBay.com. I find that weather effects
valuation of goods on eBay by influencing mood. Although most extant literature focuses on risk preferences as the mechanism thorough which mood affects economic behavior, I show that on eBay mood
alters bidding by affecting the bidder’s assessment of the monetary value of the item. This mechanism is
tested against alternative mechanisms using a longitudinal database of all bids submitted to auctions on
eBay.com for a set of five very different types of common household goods over a 3-year period. These
goods include: calculators, perfume, gift cards, computer software and apple ipods. Taking advantage
of the differences in the goods in our data set, I devise a series of empirical tests to not only identify the
average effects of weather on bidding, but tease out the mechanism though which weather effects the
bids submitted on eBay.
This paper empirically shows that on a rainy day, bidders on eBay increase their bids by about 1.5%
on average on consumer products. Rain does not seem to have any statistical effect on bidders’ WTP for
cash-like items such as gift cards. Additionally, rain has no effect on a bidder’s bid submission behavior
3
including the number of bids submitted, and how close the bid is submitted to the end of the auction.
If the weather primarily influenced WTP through altering cognitive behavior that influences bidding
strategies, or by increasing travel costs by making roads dangerous, then theory would predict that the
results should be similar for both consumer goods and cash-like substitutes. Additionally, if the weather
was decreasing the opportunity cost of bidding by increasing probability that the bidder is indoors and
in front of his or her computer, then theory would predict that weather would affect submission behavior
including the number and timing of bidders’ bids. Finally, if mood where to be affecting bids by altering
bidders’ risk preferences, then the effects of weather should be moderated by the number of other simultaneous auctions. While this is not a perfect proxy for risk, bidders should be willing to assume more
risk by bidding higher when there are fewer other auctions to turn to. Together these results indicate that
weather affects a bidders’ WTP primarily by influencing mood, which causes a revaluation of WTP for an
item.
This paper contributes to the behavioral economics literature, by detailing the effects of seasonal affect disorder, and more generally weather on bidding behavior. More specify, this is the first non experimental study, I am aware of, to focus on consumer items as opposed to investment goods such as stocks
or art in an auction setting. This difference is important; prior literature has hypothesized that seasonal
affective disorder impacts markets through mood effects on risk aversion (Kamstra et al., 2003). While
this mechanism may explain variations in bidding and valuation in art auctions or in the stock market, it
might not explain variation in consumer product markets. I instead find evidence that seasonal affective
disorder is directly altering the perceived value of the item.
Section 2 reviews the existing literature and presents my hypotheses. Section 3 describes the empirical setting and data. Section 4 describes the empirical specifications and results. And section 5 discusses
and concludes the paper.
4
2 Existing Research and Hypotheses
2.1 Weather and Mood
The use of weather as a proxy for mood is based on substantial research in neuroscience and psychology. Neuroscientists believe that sunlight and weather affect mood though the body’s level of dopamine
and melatonin: production of melatonin by the pineal gland is inhibited by light to the retina and permitted by darkness. The body usually retains a balance between dopamine and melatonin. During the
day dopamine levels are elevated and melatonin levels are down, and during the night dopamine levels
decrease as melatonin levels increase. However, during cloudy weather, the lack of sunlight causes melatonin production to rise, which decreases dopamine levels and causes a more negative mood (Rosenthal
et al., 1984; Nelson, 2005). Some researchers believe that as as much as 50% of the population experiencing some sort of seasonal depression (Dam et al., 1998). Additionally, the theoretical literature in psychology has highlighted that weather can influence mood by: (1) triggering emotional states of sadness,
anxiety, irritability, and depression (Howarth and Hoffman, 1984), (2) leading to pessimistic thinking
and apathy (Wright and Bower, 1992), (3) or causing physical discomfort which leads to negative mood
(Steers and Rhodes, 1978).
2.2 Weather and valuation
Most studies which focus on the effects of weather on valuation have done so in the context of investment
goods.2 Saunders (1993) shows that when it is cloudy in New York City the New York Stock Exchange index returns tend to be negative. This finding is further substantiated by Hirshleifer and Shumway (2003)
who look at weather effects on stock returns from 1982 to 1997 in 26 different cities where stock exchanges are located. These and a variety of other works all find a negative relationship between weather
and perceived stock returns — leading to a lower valuation.3 . DeHaan et al. (2015) extends this literature
significantly by focusing on isolating mood as the mechanism through which weather causes pessimism
in markets, and rulling out alternative mechanisms.
2 Studies have used weather as a proxy for mood in other settings including car purchases, absenteeism, voter turnout and
stock returns (Coleman and Schaefer, 1990; Markham and Markham, 2005; Lamare, 2013; Busse et al., 2015; Saunders, 1993).
3 Goetzmann and Zhu (2005), Chang et al. (2008), Dolvin et al. (2009), and Novy-Marx (2014) all also study the effects of
weather on stock returns)
5
Weather effects on auctions have also been studied both in a laboratory as well as in the real world.
Drichoutis et al. (2010) analyze the effects of mood in a second price Vickrey auction setting in a laboratory and find that subjects in a positive mood provide lower bid values than subjects in either the
negative or control mood states. Two papers which study weather effects on auctions in the real world
are Kliger et al. (2010) and De Silva et al. (2012). They both study art prices at auctions conducted in
England. They both show that sunlight had significant positive effect on the auction selling price. Presumably, the length of the day is directly linked to sunlight, which influences mood. Kliger et al. (2010)
also finds a postitive correlation between rain and auction selling price. The subjective nature of art
makes it difficult to control for item-level variation in preferences; additionally, longer days could possibly be correlated with longer auctions and increased bidding especially in the era prior to electricity.
Thus, without bid level data to test for these alternative hypothesis, neither study is unable to fully isolate
the mechanism through which weather affects WTP.
2.3 Mechanisms
Any paper which seeks to understand if weather affects valuation or other economic outcome by influencing mood must begin by exploring the non-mood explanations for weather affects on valuation.
Firstly, weather can directly alter the value of an item. For example, an umbrella is worth more when
it’s raining that when it’s clear and sunny out. Additionally, weather often alters opportunity costs. A
common saying warns, "Don’t waste a sunny day.", yet almost no one worries about waisting a rainy day.
As such, it is not unreasonable to presume that on rainy days, the opportunity cost of locating a good
deal is lower (Simonsohn, 2010). In an eBay context, on rainy days, a bidder might spend more time on
their computer monitoring auctions. This decrease in opportunity costs could lead to decrease in bids
if bidders searched for auctions with less competition, or an increase in bids if bidders get swept up in
bidding fever and revise their bids upwards.
Even if weather does alter a bidder’s mood, it is unclear how exactly mood might influence willingness to pay or valuation in the context of eBay. To understand the mechanism through which mood is
altering valuation in these settings, and the setting I study, I turn to Capra et al. (2010) who draws on the
constructed preferences hypothesis (Payne et al., 1999) under which people are assumed to construct
6
Figure 1: Mechanisms Through Which Rain Effects Submitted Bids
preferences on the spot when asked to reach a particular judgment or decision. According to this theory,
when submitting a bid, an individual constructs his or her preferences at that time. The constructive
nature of preferences means that decisions and decision processes are highly contingent upon a variety
of factors including mood. For example, a positive mood state may cause an individual to focus his or
her attention on positive attributes of an item and assign it a higher valuation (Forgas, 1995). Conversely,
consumer behavior could be at some level driven by unconscious needs to alter bad mood or promote
good mood. As such, an individual in a negative affect state might be willing to increase the amount
she is willing to pay for an item as she might belive that acquiring that item will alter her mood state
(Winslow, 1986; Billig, 1999; Isen and Geva, 1987).
Thus, I posit my first hypothesis:
Hypothesis 1 (H1): Rain affects one’s willingness to pay through affecting perceived value of the item.
2.4 Alternate Mood Mechanisms
Figure 1 details all of the ways through which weather could affect bidding behavior on eBay. In addition
to my hypothesized mechanism, mood can influence WTP in two other ways. Firstly, weather might
affect mood which influences cognitive processes, changing bidding strategies and indirectly changing
the amount bid. For example, mood could cause a bidder to become more or less competitive. In this
scenario, the value for the item is unchanged, but the value elicitation mechanism (in this case bidding)
7
may be influenced by mood. Indeed, mood has been shown to alter cognition, which, in an auction
environment, may result in an increase/reduction in deviations from the best strategy (Chepenik et al.,
2007; Maier and Watkins, 1998).
Secondly, weather could also affect mood which alters risk preferences, which affect an individual’s
WTP for some asset. This is the mechanism highlighted by most of the behavioral finance literature.
Specifically, seasonal affective disorder increases risk aversion (Kamstra et al., 2003). Thus, prices should
be positively correlated with positive weather. If buyers have higher risk aversion in bad weather they
demand higher returns for a given risk profile, and thus will pay lower prices. While in a good mood,
induced by good weather, individuals who are more risk-tolerant will be willing to pay a higher price.
To confirm our H1 hypothesis, that rain increases one’s willingness to pay through increasing perceived value of the item, I will need to test and reject these alternative mechanisms.
2.5 Mood and valuation
While the literature in psychology has detailed preference construction as a mechanism through which
mood influences valuation, the literature, is mixed on the direction of the impact of this mood mechanism on valuation. Drawing on Freudian theory, which proposes that human motivation is a result of
unconscious needs and drives, some literature proposes that consumer behavior could be at some level
driven by unconscious needs to alter bad mood or promote good mood (Winslow, 1986; Billig, 1999).
Building on this logic, the affect infusion model (Forgas, 1995), suggests that positive mood should result
in higher valuation. It hypothesizes that positive mood causes subjects to focus their attention on the
positive attributes of an item, and thus predicts a higher valuation under a positive mood and a lower
valuation under a negative mood. Alternately, the mood maintenance hypothesis (Isen and Geva, 1987)
predicts that people in a negative mood will seek to alter their negative mood and such increase their
valuation.
In this paper, all of the goods analyzed are consumer products, which are purchased immediately, but
not delivered for many days. Studies have found that people tend to overestimate the impact of future
events on their emotional lives (Gilbert et al., 1998). Additionally, individuals under negative affected
states exhibit stronger negative mood prediction biases increasing their belive that they will be stuck in
8
a negative mood state in the future (Wenze et al., 2012). Thus, I hypothesize that when bad weather
leads to a lower mood, individuals bidding on eBay will subconsciously increase their WTP, as a means
of seeking to altering their future mood, which they belive might still be negative.
Hypothesis 2 (H2): Rain will have a positive effect on willingness to pay.
3 Empirical Setting and Data
eBay serves as an ideal empirical setting to understand if and to what extent weather affects valuation,
and then to isolate if weather is altering valuation by affecting mood, and if so, through what mechanism.
Firstly, online auction sites such as eBay sell a multitude of goods, with different attributes including
some which are unlikely to affected by some mood mechanisms. This allow for comparison across goods
to help disambiguate these theories. Additionally, the mechanism through which bidders submit bids
is unlikely to be affected by “bidding fever”. In eBay auctions, bids are submitted by a proxy bidding
system. When bidding on an item, users enter the maximum amount they would be willing to pay for
the item. The seller and other bidders don’t know the maximum bid. The eBay proxy bidding system
will automatically place bids on the bidder’s behalf using the automatic bid increment amount, which
is based on the current highest bid. The system only bids as much as necessary to make sure that the
bidder remains the high bidder, or to meet the reserve price, up to the entered maximum amount. Thus,
by adding a bidder’s highest proxy bid in an auction to the posted shipping price, I am able to construct a
variable which represents an individual’s willingness to pay in a particular auction for a particular good.4
Secondly, eBay is the largest online auction market, with over $6,000 million in marketplace sales in 2015,
and is itself an important market to analyze.
Bid-level eBay data was purchased from Advanced Ecommerce Research Systems (AERS). For each
auction several characteristics are observable: the auction title, subtitle, and category listing. Additionally, I observe the auction starting time and starting price as well as ending time and ending price. A
unique identifier for each seller is also observable. I also observe the seller’s total feedback score, if the
auction included a Buy It Now option and the Buy It Now price. Finally, I observe several auction-level
4 Brown et al. (2007) show that bidders on eBay often fail to properly account for shipping. All of the results presented are
robust to specifications that do not include shipping.
9
indicator variables that denote if the auction was specially featured in any way. For example, the seller
could pay extra for the auction to have a bolded title or high-resolution pictures. For each auction, I observe every bid amount submitted to eBay’s proxy bidding system by each bidder. Bidders are identified
by a unique bidder identifier. The bidder’s total feedback score and the exact time of the bid is recorded.
I also observe the current price of the auction when each bid is submitted.
I restrict my analysis to a group of carefully selected items covering a wide variety of types of items
available on eBay. These include: TI-83 Plus calculators, second-generation iPod Nanos, Giorgio Armani
Acqua Di Gio for men 3.4 oz sets of perfume, Target gift cards, and Windows Vista Home Premium and
Ultimate operating systems. These goods were selected based on their popularity and homogeneous nature. The item space includes software, hardware, luxury goods and non-luxury goods. It also includes
goods that depreciate in value over time as well as well as cash-substitutes that experience no depreciation. The data includes both used and new items. The AERS data covers all transactions for the above
items for auctions ending from January 1, 2008, to December 31, 2010. Due to the multiple-year nature
of the data, significant variation in supply and demand as well as they types of buyers and sellers may
occur even within a product group. This variation will need to be addressed econometrically.
Historical weather data was obtained from Weather Underground (www.wundergound.com) using a
web-scraper program. Data obtained from weather underground is preferable to that supplied by the National Climatic Data Center (NCDC) as it supplements the 22,000 weather stations supported by NCDC
with 18,000 additional personal weather stations. This allows for weather to be observed at the zip code
level as opposed to the weather station level. Observations include temperature, dew point, pressure,
visibility, wind speed, precipitation cloud cover, and weather events such as hail, thunderstorms, or tornados. The historical weather data was matched to the bid-level eBay data by date and zip code. Zip
code data is however only observable for winners of auctions, but by matching based on unique bidder
ID I am able to match 42% of the total bidders to zip codes, as 42% of bidders win at least one auction in
my sample.5
This matching method relies on the assumption that the bidder location remains fixed between the
5 See Appendix B for a comparison of bidding behavior between eventual winners and those bidders who never win an
auction. Eventual winners on average bid much lower than those who never win, and also bid much further from the end of the
auction. This indicates that eventual winners are much more selective about what auctions they participate in. Additionally,
eventual winners have statistically lower variances in their bidding behavior.
10
moments when he or she bids and does not win, and when he or she wins an auction. Another issue
would be if purchases were shipped to a location which differs dramatically from where it is purchased.
On average, a non-winning bid is matched, with zip code information, to a winning bid that is within
79.67 days. The median match is only 2.69 days apart, and the 75th percentile match is only 59.89 days.
While I must assume that the bidder did not move between when I observe his zip code and when he bid,
this assumption seems resonable since 75% of the matches are less than two months apart. When there
are multiple zip-codes per bidder, the match that is closest in days is used. This matching methodology
is necessary for the analysis, but it is important to realize that it is a potential source of error. Only 11.9%
of Americans moved during 2007 and 2008 and 12.5% of Americans moved during 2009. Additionally,
most of these moves were within the same county.6 These statistics combined with the short number of
days between most of the matches makes the assumption of fixed bidder location even more justifiable.
A more conservative approach would be to run the regressions at an auction level using only the winning
bid price. However, since bidders on eBay usually bid across many auctions until they win, this would
reduce my sample to about one bid per individual. This would prohibit me from taking advantage of
individual fixed effects regressions. Additionally, my question of interest is at the bid level not the auction
level. I am not interested in if weather affects the price of a good sold on eBay, but if it predicts variation
in a bidder’s willingness to pay.
To help get a sense of the amount of variation in bidding amounts, I present some summary statistics
in table 1. These statistics will prove insightful when investigating if the calculated effect of weather on
bid amounts is economically meaningful. The statistics are calculated at the auction level. For statistics
separated by product see appendix A.
In the final data set, attention is restricted to successfully completed auctions with known shipping
costs and normal durations. Furthermore, in an attempt to limit the product space auctions with uncommon condition values are removed. More precisely, item quality can be broken into three categories:
new, used and unknown. The analysis focuses on the most common condition per product: new for perfume, gift cards, and Windows; used for calculators and iPods. Minority conditions and broken items are
never considered. Additionally, the top and bottom 1% of bids by item group are trimmed. The top 1%
6 http://online.wsj.com/article/SB10001424052748704879704575236533316039428.html
11
Table 1: Per-Auction summary statistics
Statistic
Bid + Shipping
Bid
Hours from bid to auction end
Seller feedback
Buyer feedback
Winning price
Num bid per auction
Num bid per person
High Bid per person
Low bid per person
Hours from last bid per person
Auction with bold fee
Auction with picture fee
Auction with schedule fee
Bids while raining
N
Mean
St. Dev.
Min
Median
Max
39,113
39,550
39,550
39,550
39,550
39,113
39,550
39,550
39,113
39,113
39,550
39,550
39,550
39,550
37,671
73.06
68.31
31.94
3,966.56
401.22
95.31
12.24
1.76
77.75
70.06
31.83
0.02
0.10
0.10
0.27
164.38
163.55
29.85
21,258.21
815.42
81.73
7.50
0.74
256.87
140.16
29.25
0.15
0.31
0.30
0.34
5.73
0.99
0.00
−999.00
−498.50
7.99
1.00
1.00
6.59
2.53
0.00
0
0
0
0.00
49.28
45.15
23.65
119.00
183.33
66.66
11.00
1.60
51.48
46.58
24.59
0
0
0
0.11
23,630.55
23,617.56
232.68
372,020.00
56,055.67
1,025.00
90.00
11.50
35,373.95
23,079.85
229.97
1
1
1
1.00
Notes: This table reports auction level summary statistics. Where noted as ’per person’, individual per auction statistics were calculated first,then auction level averages of those are reported.
of bids are removed to get rid of shill bids, which is the practice by sellers of bidding a high price on their
own item when the auction might close at a low price, so they can repost the auction. The bottom 1%
of bids are removed to eliminate false WTP generated by the practice of bidding a very low amount on
an auction to more easily follow the auction moving forward. eBay later added a Follow Auction which
greatly reduced this behavior.
4 Empirical Analysis and Results
To explore the relationship between weather and WTP, I regress exogenous variation in weather on the
highest bid submitted by a bidder in an auction on eBay.com. I control for bidder heterogeneity, time
effects, and auction level heterogeneity. I use individual, item, and time fixed effects. This means that
the effect is measured from variation in an individual’s bid across different auctions for the same item
within the same month. I do not analyze the effect of temperature, as temperature is not a monotonic
measure of mood. For example, mood may be higher on a 70-degree day than on a 50-degree day, but
mood is probably not higher on a 95-degree day than on a 70-degree day.
12
I begin with a regression of the following form:
B IDVALUEi pt z = β0 + β1 R AIN t z + β2 Xi t + δi + γp + ζt + ²i pt z
(1)
where the dependent variable B IDVALUEi pt z is the bid amount plus shipping cost placed by individual
i on product p at year-month t in zip code z. The explanatory variable of interest is an indicator of rain
at time t in zip code z.7 Xi t is a set of auction level covariates including: the auction’s minimum bid, the
number of previous bids submitted at the time of the focal bid, an indicator for the weekend, the number
of simultaneous auctions, the seller’s feedback, an indicator if the auction had a reserve interacted with
the reserve price, an indicator if the auction had pictures, and an indicator if the auction paid extra for
bold text in the listing. δi is an individual fixed effect, γp is an item fixed effect, and ζt is a month-year
fixed effect.
The regressor of interest (rain) varies only at the zip code-day level, but the dependent variable (bid
+ shipping) varies at the individual-zip code-day level. Thus, I expect that there is substantial serial
correlation in the residual within zip code. To solve this I replace the standard OLS variance estimator
with a robust cluster-variance estimator with clustering by zip code.8 More conservatively, all regressions
were also run with clusters at the county level, with no change in what variables are significant.
This analysis necessitates the use of a multi-month, multi-year dataset. This introduces complications as significant variation may occur across products, conditions, and categories during the sample.
Any specification that fails to include month-year fixed effects is inherently flawed as there are unobserved time trends in bidding that are correlated with season and by extension, weather. For example,
the period between Thanksgiving and Christmas is regarded as the biggest shopping time of the year.
In fact, aiming to boost its standing as a holiday shopping destination, eBay launched a holiday season
marketing campaign in 2009, which it renewed in 2010.9 This time of year also tends to be characterized
by more cloudy, snowy, and rainy weather for much of the country. This unobserved correlation between
weather, time, and unobserved demand and supply shocks must therefore be controlled for using month
7 In table 2 in C I use the bid without shipping as the dependent variable, and the results are nearly identical.
8 I am actually concerned by both correlation within zip code as well as over time, I aggregate only at the group level as
recommended by Arellano (1987) and Angrist and Pischke (2009).
9 http://online.wsj.com/article/SB10001424052748704746304574505543900212118.html
13
fixed effects. Additionally, I use month-year fixed effects as there are also unobserved shocks which vary
by month and year. For example, Windows 7, the successor to Windows Vista, was released on October
22, 2009.
I am still concerned about unobserved individual treatment-level heterogeneity. I am concerned that
people select themselves into the environment in which they live. Hypothetically, it’s possible that people who love cloudy weather and rain and shun material items – reducing their valuation of items on eBay
– choose to live in the Pacific Northwest, while those whose moods are most profoundly affected by sunlight and who place higher value on material goods select into living in southern California. This would
cause an unobserved correlation between weather and bidding across geographic region. To correct for
this possibility, I add individual fixed effects to the model. While the above story could be corrected for
by using zip code-level fixed effects, I decide to follow the even more conservative methodology and use
individual fixed effects which adjust for the additional possibility that there is some additional unobserved variable correlated with individual characteristics and willingness to pay. The results using zip
code-level fixed effects are nearly identical to those with individual fixed effects. I am now only using
variation within an individual, month, year, and item to drive the results.
To better understand what kinds of bidders drive the variation in my final sample, let us quickly
explore how often bidders bid on and win a particular item in a month. Most of the bidders in my final
dataset are individuals who desire to obtain one unit of the good demanded. Of the bidders, for whom I
have weather data for, 36.5% who bid on a particular item in a given month fail to win that item in that
month. While 58.1% of bidders win exactly one item in the month they bid on it, and the remaining 5.3%
of bidders end up winning multiple copies of the item in the month they bid on it. On average individuals
in my sample bid in 2.3 auctions per month for a given item. While, the 5.3% of bidders who win multiple
copies of the same item bid in 8.2 auctions per month.
4.1 Rain effects on consumer products
To empirically isolate the different channels though which the weather, namely rain, affects willingness
to pay, I designed a series of empirical tests that take advantage of the different items in the dataset. I
first note that Target gift cards are fundamentally different from the other items in the dataset in that
14
their value is common to all bidders. In fact, having an associated dollar amount effectively induces a
valuation. Since a $100 Target gift card is worth exactly $100 (minus some amount due to lack of spending
flexibility), it is unlikely that the perceived value of the gift card will be altered by mood. If weather affects
the willingness to pay for a Target gift card, it will likely do so by altering the way subjects decide what to
bid. The results reported in table 2 omit target gift cards from the analysis.
Table 2: The Effect of Rain on Willingness to Pay, Consumer Goods
Dependent variable:
Bid + Shipping
Rain Indicator
Opening Level Covariates
Month-Year FE
Item FE
Bidder FE
Observations
(1)
(2)
(3)
2.145∗∗∗
(0.392)
1.824∗∗
(0.898)
1.167∗
(0.665)
Yes
No
No
No
47,110
Yes
Yes
Yes
No
47,110
Yes
Yes
Yes
Yes
47,110
Notes: Covariates include: the minimum bid, the bidder’s previous bid count, weekend
indicator, number of simultaneous auctions, seller feedback, indicator for reserve, indicator for bold auction title, indicator for picture in auction. All models are OLS with robust standard errors clustered at the zip code level. Significance indicators: p ≤ 0.10 : ∗,
.p ≤ 0.05 : ∗∗ and p ≤ .01 : ∗ ∗ ∗.
Table 2 shows that rain has an effect on bids submitted on eBay on non cash-like items, and that
this effect is positive confirming hypothesis H2. This effect could be due to mood or opportunity costs,
which I will study in section 4.4. Model (1) of table 2 includes only the vector of covariates Xi t . Rain
on the day of the bid corresponds to a $2.14 increase in the average bid. This implies that rain has a
significantly positive effect on a bidder’s valuation of a good on eBay. In model (2) I add in month-year
fixed effects and item-level fixed effects. Including time fixed effects and item fixed effects does not have
any meaningful effects on the outcome of the regression. In model (3) I run the preferred specification
which uses individual bidder fixed effects in addition to the the month-year fixed effects and item fixed
effects. Here the coefficient is slightly smaller with rain on the day of the bid corresponding to an increase
in average bid of $1.16.
15
4.2 Rain effects on giftcards
I next run the same regression, but limit the sample to only Target gift cards. This will help to establish
the mechanism through which weather and by proxy mood influence willingness to pay (H1). If weather
has a similar effect on bidding for Target gift cards as it did for non-Target gift cards, it is more likely
that weather does not have any effect on the perceived value of the item. Table 3 shows that there is no
significant difference between willingness to pay on days when it rains and days when it does not rain.
Table 3: The Effect of Rain on Willingness to Pay, Gift Cards
Dependent variable:
Bid
(1)
(2)
(3)
Rain Indicator
−0.484
(1.448)
0.023
(2.752)
0.176
(1.534)
Opening Level Covariates
Month-Year FE
Bidder FE
Observations
Yes
No
No
30,579
Yes
Yes
No
30,579
Yes
Yes
Yes
30,579
Notes: Covariates include: the minimum bid, the bidder’s previous bid count, weekend
indicator, number of simultaneous auctions, seller feedback, indicator for reserve, indicator for bold auction title, indicator for picture in auction. All models are OLS with robust standard errors clustered at the zip code level. Significance indicators: p ≤ 0.10 : ∗,
.p ≤ 0.05 : ∗∗ and p ≤ .01 : ∗ ∗ ∗.
4.3 Rain effects by number of simultaneous auctions
Next, I investigate how the number of simultaneous auctions for an identical good mediates the effect
of mood on bidding. To do so, I rely on the assumption that when bidding on eBay, individuals who are
more risk adverse will increase their willing to pay for an item when there are fewer alternative auctions
he or she can bid in if they fail to win the item in the focal auction. While the setting and types of goods
studied in this paper are less likely to be influenced by risk aversion, there is limited evidence that risk
aversion does influence bidding on eBay at least in auctions with buy it now options – which where were
excluded from this analysis ().
16
Bid+Shipping
(demeaned by bidderID and month−year)
Figure 2: WTP as a function of number of overlapping auctions by rain
2
1
No Rain
Rain
0
−1
−5
0
5
Number of Overlaping Auctions
(demeaned by bidderID and month−year)
Note: The figure shows bidders highest submitted bid + shipping plotted against the number of simultaneous auctions for that product. Both the highest bid and the number of overlapping auctions have been
demeaned by unique bidderID and month. Thus, the plots can be thought of including month-year-bidder
fixed.
In figure 2 I plot the relationship between number of overlapping bids and the bidders WTP in the
auction by if the bid was submitted on a rainy day. To eliminate systematic bidder level differences as
well as systematic increases in both number of overlapping auctions as well as WTP due to seasonal
effects, I demeaned the number of overlapping auctions and the bid plus shipping amount by unique
bidderID, item, and month-year. We observe a negative relationship between number of overlapping
auctions and WTP, as the more simultaneous auctions, the more chances the bidder has to win the item.
The change in WTP with number of overlapping auctions is not statistically different on rainy and not
rainy days. While not a perfect test, this provides some additional support that mood is not influencing
WTP on eBay by altering risk preferences.
4.4 Alternative mechanisms
To fully confirm or reject if weather is increasing bids on eBay through mood (H1) I must run a series of
robustness check to rule out alternative mechanisms. It’s possible that on days with worse weather (e.g.
rain) a bidder would rather be indoors instead of outside. Being stuck indoors allows the bidder to follow
17
an auction more closely which causes him to get caught up in bidding fervor and increase his bid. To
check if this is a viable alternative hypothesis, I devise two robustness checks. First, I calculate the time
from when the bid took place to the end of the auction, as I hypothesize that if a bidder was following
an auction more closely, he would wait longer to submit his bid. If this alternate story were true, I also
expect bidders to submit more bids per auction on days when they are stuck indoors.
Model (1) of table 4 regresses the time to the end of the auction in seconds on a rainy day indicator
and auction-level covariates. I include individual, and item-month-year fixed effects just as in the main
specification. Rain does not seem to have any significant effect on how close to the end of the auction a
bid was submitted, as I am unable to reject the null hypothesis that rain influences the timing of submitted bids. Model (2) regresses the number of bids submitted per day on a rainy day indicator. Again, the
indicator for a rainy day is not significantly different from zero, thus I can not reject the null hypothesis
that rain does not influence the number of bids submitted. These robustness checks seem to indicate
that the effect of weather on bidding is probably not only due to bidders paying more attention to an
auction.
Table 4: The Effect of Rain on Timing/Quantity of Bids, Non Gift Cards
Dependent variable:
Rain Indicator
Opening Level Covariates
Month-Year FE
Item FE
Bidder FE
Observations
Min To End
Bid/Day
(1)
(2)
35.993
(34.487)
0.029
(0.030)
Yes
Yes
Yes
Yes
47,110
Yes
Yes
Yes
Yes
34,966
Notes: Covariates include: the minimum bid, the bidder’s previous bid count, weekend
indicator, number of simultaneous auctions, seller feedback, indicator for reserve, indicator for bold auction title, indicator for picture in auction. All models are OLS with robust standard errors clustered at the zip code level. Significance indicators: p ≤ 0.10 : ∗,
.p ≤ 0.05 : ∗∗ and p ≤ .01 : ∗ ∗ ∗.
18
5 Discussion and Conclusion
I find that a rainy day is associated with a bidder on eBay submitting bids on consumer items that are
about 1.5% higher on average than those submitted on non-rainy days. This relationship only holds
for general consumer items. There is no statistically significant relationship between rain on the day
a bid is submitted and the bid submitted for Target gift cards, which can be considered a type of cash
substitute. Additionally, the number of simultaneous auctions for the same good, does not seem to have
a differentiated moderating effect on WTP by rain. Finally, rain does not seem to have any effect on either
the number of bids submitted on that day, or how close to the end of the auction the last bid on a given
auction was submitted.
These empirical tests help tease out the mechanism through which weather and possibly mood affect
bidders’ willingness to pay for items on eBay. The products I analyze can be separated into two general
categories: consumer goods, which can generally be considered private value goods, and cash-like substitutes, which can be considered public value goods. This differentiation is important as it allows me
to unpack the mechanism through which mood effects willingness to pay. The literature in psychology
has highlighed that mood can directly affect an individual’s perceived value of an item. The literature
has also detailed that mood influences cognitive processes, changing bidding strategies and indirectly
changing the amount bid. If mood were to affect cognitive processes more generally, it seems likely that
the effect would also be seen on cash-like substitutes. If mood were influencing valuation by altering risk
preferences, I would expect individuals who are more risk adverse to increase their willingness to pay for
an item when there are fewer alternative auctions he or she can bid in if they fail to win the item in the
focal auction. Finally, to help alleviate concerns that the effects of the weather on bidding are not due to
mood at all, but instead to changes in opportunity costs of bidding, I show that weather has no effect of
the timing and number of bids submitted in an auction.
Taken all together, the evidence presented seems to suggest that the weather, in particular rain, does
have a significant yet small positive effect on bidding behavior on eBay. It further suggests that the mechanism through which weather affects bidding behavior on eBay is by altering mood, which alters a bidder’s perceived valuation of an item. Our paper sheds light on the economic importance of
It is important to note the limitations of the analysis in this paper. One very important limitation is
19
that due to my method of matching, my sample is limited to bidders who have won at least one of my
analyzed items on eBay from January 2008 through December 2010. This means that I have a selected
sample of winners compared to the average bidder on eBay.
My results are potentially relevant for understanding how preferences are constructed and how individuals assign economic value to goods. Specifically, these results lend support for the constructed
preferences hypothesis (Payne et al., 1999) under which people are assumed to construct preferences
on the spot when asked to reach a particular judgment or decision. They further demonstrate that the
effects of mood can help elucidate consistent revaluation of consumer products by individuals. These
finding also have implications for understanding seeming bias in managerial decisions such as hiring
and promotions where preferences are constructed on the spot.
20
References
Bachkirov, Alexandre A (2015). “Managerial decision making under specific emotions”. Journal of Managerial Psychology 30.7, pp. 861–874.
Bassi, Anna, Riccardo Colacito, and Paolo Fulghieri (2013). “’O sole mio: an experimental analysis of
weather and risk attitudes in financial decisions”. Review of Financial Studies, hht004.
Billig, Michael (1999). “Commodity fetishism and repression: reflections on Marx, Freud and the psychology of consumer capitalism”. Theory & Psychology 9.3, pp. 313–329.
Brief, Arthur P and Howard M Weiss (2002). “Organizational behavior: Affect in the workplace”. Annual
review of psychology 53.1, pp. 279–307.
Brown, Jennifer, Tanjim Hossain, and John Morgan (2007). “Shrouded attributes and information suppression: Evidence from the field”. The Quarterly Journal of Economics, pp. 859–876.
Busse, Meghan R et al. (2015). “The psychological effect of weather on car purchases”. The Quarterly
Journal of Economics 130.1, pp. 371–414.
Capra, C Monica, Kelli F Lanier, and Shireen Meer (2010). “The effects of induced mood on bidding in
random nth-price auctions”. Journal of Economic Behavior & Organization 75.2, pp. 223–234.
Chang, Shao-Chi et al. (2008). “Weather and intraday patterns in stock returns and trading activity”. Journal of Banking & Finance 32.9, pp. 1754–1766.
Chepenik, Lara G, Lauren A Cornew, and Martha J Farah (2007). “The influence of sad mood on cognition.” Emotion 7.4, p. 802.
Coleman, Daniel F and Norbert V Schaefer (1990). “Weather and absenteeism”. Canadian Journal of
Administrative Sciences/Revue Canadienne des Sciences de l’Administration 7.4, pp. 35–42.
Dam, H, Klaus Jakobsen, and E Mellerup (1998). “Prevalence of winter depression in Denmark”. Acta
Psychiatrica Scandinavica 97.1, pp. 1–4.
De Silva, Dakshina G, Rachel AJ Pownall, and Leonard Wolk (2012). “Does the sun ‘shine’on art prices?”
Journal of Economic Behavior & Organization 82.1, pp. 167–178.
DeHaan, Ed, Joshua M Madsen, and Joseph D Piotroski (2015). “Do weather-induced moods affect the
processing of earnings news?” Available at SSRN.
21
Denissen, Jaap JA et al. (2008). “The effects of weather on daily mood: A multilevel approach.” Emotion
8.5, p. 662.
Dolvin, Steven D, Mark K Pyles, and Qun Wu (2009). “Analysts get SAD too: The effect of seasonal affective
disorder on stock analysts’ earnings estimates”. The Journal of Behavioral Finance 10.4, pp. 214–225.
Drichoutis, Andreas, Rodolfo Nayga, and Stathis Klonaris (2010). “The effects of induced mood on preference reversals and bidding behavior in experimental auction valuation”.
Forgas, Joseph P (1995). “Mood and judgment: the affect infusion model (AIM).” Psychological bulletin
117.1, p. 39.
Gilbert, Daniel T et al. (1998). “Immune neglect: a source of durability bias in affective forecasting.” Journal of personality and social psychology 75.3, p. 617.
Goetzmann, William N and Ning Zhu (2005). “Rain or shine: where is the weather effect?” European
Financial Management 11.5, pp. 559–578.
Hannak, Aniko et al. (2012). “Tweetin’in the Rain: Exploring Societal-Scale Effects of Weather on Mood.”
ICWSM. Citeseer.
Hirshleifer, David and Tyler Shumway (2003). “Good day sunshine: Stock returns and the weather”. The
Journal of Finance 58.3, pp. 1009–1032.
Howarth, Edgar and Michael S Hoffman (1984). “A multidimensional approach to the relationship between mood and weather”. British Journal of Psychology 75.1, pp. 15–23.
Isen, Alice M and Nehemia Geva (1987). “The influence of positive affect on acceptable level of risk: The
person with a large canoe has a large worry”. Organizational Behavior and Human Decision Processes
39.2, pp. 145–154.
Isen, Alice M, Andrew S Rosenzweig, and Mark J Young (1991). “The influence of positive affect on clinical
problem solving”. Medical Decision Making 11.3, pp. 221–227.
Johnson, Eric J and Amos Tversky (1983). “Affect, generalization, and the perception of risk.” Journal of
personality and social psychology 45.1, p. 20.
Kamstra, Mark J, Lisa A Kramer, and Maurice D Levi (2003). “Winter blues: A SAD stock market cycle”.
The American Economic Review 93.1, pp. 324–343.
Kliger, Doron et al. (2010). “Auction Prices and the Weather: New Evidence from Old Masters”.
22
Kööts, Liisi, Anu Realo, and Jüri Allik (2011). “The influence of the weather on affective experience”.
Journal of individual differences.
Kuhnen, Camelia M and Brian Knutson (2011). “The influence of affect on beliefs, preferences, and financial decisions”. Journal of Financial and Quantitative Analysis 46.03, pp. 605–626.
Lamare, J Ryan (2013). “Mobilization and Voter Turnout: Should Canvassers Worry about the Weather?”
PS: Political Science & Politics 46.03, pp. 580–586.
Lazarus, Richard S (1991). “Cognition and motivation in emotion.” American psychologist 46.4, p. 352.
Lerner, Jennifer S, Deborah A Small, and George Loewenstein (2004). “Heart strings and purse strings
carryover effects of emotions on economic decisions”. Psychological science 15.5, pp. 337–341.
Loewenstein, George F, Elke U Weber, et al. (2001). “Risk as feelings.” Psychological bulletin 127.2, p. 267.
Loewenstein, George and Jennifer S Lerner (2003). “The role of affect in decision making”. Handbook of
Affective Science 619.642, p. 3.
Maier, Steven F and Linda R Watkins (1998). “Cytokines for psychologists: implications of bidirectional
immune-to-brain communication for understanding behavior, mood, and cognition.” Psychological
review 105.1, p. 83.
Markham, Steven E and Ina S Markham (2005). “Biometeorological effects on worker absenteeism”. International journal of biometeorology 49.5, pp. 317–324.
Nelson, Randy J (2005). An introduction to behavioral endocrinology . Sinauer Associates.
Novy-Marx, Robert (2014). “Predicting anomaly performance with politics, the weather, global warming,
sunspots, and the stars”. Journal of Financial Economics 112.2, pp. 137–146.
Payne, John W et al. (1999). “Measuring constructed preferences: Towards a building code”. Elicitation of
Preferences. Springer, pp. 243–275.
Raghunathan, Rajagopal and Michel Tuan Pham (1999). “All negative moods are not equal: Motivational
influences of anxiety and sadness on decision making”. Organizational behavior and human decision
processes 79.1, pp. 56–77.
Rosenthal, Norman E et al. (1984). “Seasonal affective disorder: a description of the syndrome and preliminary findings with light therapy”. Archives of general psychiatry 41.1, pp. 72–80.
23
Saunders, Edward M (1993). “Stock prices and Wall Street weather”. The American Economic Review 83.5,
pp. 1337–1345.
Schwarz, Norbert, Herbert Bless, and Gerd Bohner (1991). “Mood and persuasion: Affective states influence the processing of persuasive communications”. Advances in experimental social psychology 24,
pp. 161–199.
Seo, Myeong-Gu, Lisa Feldman Barrett, and Jean M Bartunek (2004). “The role of affective experience in
work motivation”. Academy of Management Review 29.3, pp. 423–439.
Simonsohn, Uri (2010). “Weather to go to college”. The Economic Journal 120.543, pp. 270–280.
Smith, Craig A and Phoebe C Ellsworth (1985). “Patterns of cognitive appraisal in emotion.” Journal of
personality and social psychology 48.4, p. 813.
Steers, Richard M and Susan R Rhodes (1978). “Major influences on employee attendance: A process
model.” Journal of Applied psychology 63.4, p. 391.
Tan, ESH, D Sander, and KR Scherer (2009). “The Oxford Companion to Emotion and the Affective Sciences”. Decision making In D. Sander and KR Scherer (Eds.) Pp. 111–113.
Tsai, Wei-Chi, Chien-Cheng Chen, and Hui-Lu Liu (2007). “Test of a model linking employee positive
moods and task performance.” Journal of Applied Psychology 92.6, p. 1570.
Wegener, Duane T, Richard E Petty, and Stephen M Smith (1995). “Positive mood can increase or decrease message scrutiny: the hedonic contingency view of mood and message processing.” Journal
of personality and social psychology 69.1, p. 5.
Wenze, Susan J, Kathleen C Gunthert, and Ramaris E German (2012). “Biases in affective forecasting
and recall in individuals with depression and anxiety symptoms”. Personality and Social Psychology
Bulletin 38.7, pp. 895–906.
Winslow, EG (1986). “Keynes and Freud: Psychoanalysis and Keynes’s Account of the “Animal Spirits” of
Capitalism”. Social Research, pp. 549–578.
Wright, William F and Gordon H Bower (1992). “Mood effects on subjective probability assessment”.
Organizational behavior and human decision processes 52.2, pp. 276–291.
24
A Summary Statistics by Product
B Eventual Winners vs. Non Winners
C Additional Empirical Tests
In table 2 I rerun the same empirical specification as in table 2, but I use only the bid amount without
shipping as the dependent variable.
Table 8 shows the main specification for each item individually. I present the items together using
an item FE in the main text. The positive effect on price does not seem to hold for TI-83 calculators.
However, there are very few observations for calculators compared to the other items, so I am hesitant to
draw strong conclusions from the results.
25
Table 5: Auction summary statistics by Product
Statistic
Target Gift Cards
Bid + Shipping
Hours from bid to auction end
Seller feedback
Buyer feedback
Winning price
Num bid per auction
Num bid per person
Bids while raining
Windows Software
Bid + Shipping
Hours from bid to auction end
Seller feedback
Buyer feedback
Winning price
Num bid per auction
Num bid per person
Bids while raining
Apple iPod
Bid + Shipping
Hours from bid to auction end
Seller feedback
Buyer feedback
Winning price
Num bid per auction
Num bid per person
Bids while raining
TI83 Calculator
Bid + Shipping
Hours from bid to auction end
Seller feedback
Buyer feedback
Winning price
Num bid per auction
Num bid per person
Bids while raining
Giorgio Armani Cologne
Bid + Shipping
Hours from bid to auction end
Seller feedback
Buyer feedback
Winning price
Num bid per auction
Num bid per person
Bids while raining
N
Mean
St. Dev.
Min
Median
Max
8,001
8,024
8,024
8,024
8,001
8,024
8,024
8,002
100.99
47.49
429.78
569.29
135.55
10.31
1.41
0.25
108.55
40.18
1,509.05
931.68
143.98
6.16
0.50
0.26
5.73
0.001
−4
−2.00
8.05
1
1.00
0.00
50.75
39.81
125
421.18
71.00
9
1.27
0.20
761.25
232.68
36,429
56,055.67
1,025.00
76
9.50
1.00
17,208
17,535
17,535
17,535
17,208
17,535
17,535
16,392
86.56
33.17
2,129.82
488.90
111.38
12.70
1.84
0.29
228.27
26.16
14,214.34
955.11
52.03
7.96
0.77
0.35
12.17
0.00
−999
−498.50
25.25
1.00
1.00
0.00
72.43
28.80
75
198.00
100.00
12.00
1.67
0.13
23,630.55
213.98
372,020
43,408.00
368.99
90.00
10.50
1.00
5,100
5,148
5,148
5,148
5,100
5,148
5,148
4,952
50.06
16.53
18,504.87
185.29
67.76
14.95
2.06
0.24
32.91
19.08
48,896.48
512.91
24.64
7.93
0.84
0.36
7.21
0.0003
−1
−491.00
7.99
1.00
1.00
0.00
45.81
9.80
138
77.15
63.99
14.00
1.90
0.00
984.99
165.95
273,871
9,281.33
193.50
55.00
11.00
1.00
1,730
1,746
1,746
1,746
1,730
1,746
1,746
1,712
37.51
7.17
1,886.60
343.83
47.80
11.20
1.80
0.27
13.43
10.73
5,766.04
499.01
11.62
6.87
0.73
0.39
8.30
0.001
−2
0.00
10.25
1
1.00
0.00
36.68
3.33
129
113.53
47.00
10
1.67
0.00
310.50
149.28
67,444
4,085.50
96.00
45
7.50
1.00
7,074
7,097
7,097
7,097
7,074
7,097
7,097
6,613
33.89
28.59
2,469.44
165.32
42.17
11.55
1.74
0.27
70.43
23.32
9,766.10
277.54
6.80
6.77
0.65
0.36
10.40
0.00
−2.00
−106.33
20.50
1.00
1.00
0.00
31.82
24.44
493.00
103.20
41.59
11.00
1.62
0.00
5,790.30
174.75
193,957.00
9,028.57
91.60
49.00
11.50
1.00
Notes: This table reports auction level summary statistics for each product in the data. Where noted as
’per person’, individual per auction statistics were calculated first,then auction level averages of those are
reported.
26
Table 6: Non winners vs. eventual winners by rain
Never
Winner
mean:
X̄ C T L
Eventual
Winner
mean:
X̄ T RT
Difference
In Means
p-value
Target Gift Card
Number of Observations
Avg Bid
Avg Bid + Shipping
Bid/Amazon Price
Min to Auction End
25125
123.78
124.36
0.88
2574.81
30688
91.87
92.53
0.60
3175.04
-31.91 (1.08)
-31.83 (1.08)
-0.28 (0.00)
600.23 (24.65)
<0.001
<0.001
<0.001
<0.001
***
***
***
***
Windows Software
Number of Observations
Avg Bid
Avg Bid + Shipping
Bid/Amazon Price
Min to Auction End
55914
105.50
110.94
0.34
1571.67
27978
69.89
75.51
0.24
2731.15
-35.61 (0.37)
-35.43 (0.37)
-0.10 (0.00)
1159.48 (20.97)
<0.001
<0.001
<0.001
<0.001
***
***
***
***
Apple iPod
Number of Observations
Avg Bid
Avg Bid + Shipping
Bid/Amazon Price
Min to Auction End
26826
54.80
59.97
0.81
519.98
9602
40.30
45.63
0.64
1189.92
-14.50 (0.29)
-14.34 (0.30)
-0.18 (0.00)
669.94 (23.06)
<0.001
<0.001
<0.001
<0.001
***
***
***
***
TI 83 Calculator
Number of Observations
Avg Bid
Avg Bid + Shipping
Bid/Amazon Price
Min to Auction End
7376
37.60
43.27
0.91
215.30
3055
31.01
36.18
0.74
561.14
-6.59 (0.31)
-7.09 (0.31)
-0.17 (0.01)
345.84 (25.59)
<0.001
<0.001
<0.001
<0.001
***
***
***
***
Armani Perfume
Number of Observations
Avg Bid
Avg Bid + Shipping
Bid/Amazon Price
Min to Auction End
15839
31.43
37.61
0.79
863.42
6529
23.55
29.63
0.62
2147.45
-7.88 (0.15)
-7.98 (0.15)
-0.17 (0.00)
1284.03 (34.60)
<0.001
<0.001
<0.001
<0.001
***
***
***
***
Notes: This table reports means and standard errors for bidding behavior for each item in the dataset for never-winners and
eventual winners. Observations are at the individual-auction level. Only the highest bid for an individual in each auction is
kept. Reported p-values are the for two-sided t-tests of the null hypothesis of no difference in means across groups. In the
bottom panel, standard errors are clustered at the employer level. Significance indicators: p ≤ 0.10 :† ,p ≤ 0.05 : ∗, p ≤ 0.01 : ∗∗
and p ≤ .001 : ∗ ∗ ∗.
27
Table 7: Willingness to Pay - Consumer Goods, No Shipping
Dependent variable:
Bid
Rain Indicator
Opening Level Covariates
Month-Year FE
Item FE
Bidder FE
Observations
(1)
(2)
(3)
2.056∗∗∗
(0.386)
1.716∗
(0.889)
1.091∗
(0.644)
Yes
No
No
No
47,110
Yes
Yes
Yes
No
47,110
Yes
Yes
Yes
Yes
47,110
Notes: Covariates include: the minimum bid, the bidder’s previous bid count, weekend
indicator, number of simultaneous auctions, seller feedback, indicator for reserve, indicator for bold auction title, indicator for picture in auction. All models are OLS with robust standard errors clustered at the zip code level. Significance indicators: p ≤ 0.10 : ∗,
.p ≤ 0.05 : ∗∗ and p ≤ .01 : ∗ ∗ ∗.
Table 8: The Effect of Rain on Willingness to Pay - By Item
Dependent variable:
Bid + Shipping
calculator
perfume
Windows
ipod
targetcard
(1)
(2)
(3)
(4)
(5)
Rain Indicator
1.015
(0.819)
1.180
(0.897)
−0.120
(0.769)
0.319
(0.515)
0.176
(1.534)
Opening Level Covariates
Month-Year FE
Bidder FE
Observations
Yes
Yes
Yes
27,959
Yes
Yes
Yes
9,579
Yes
Yes
Yes
3,050
Yes
Yes
Yes
6,522
Yes
Yes
Yes
30,579
Notes: Covariates include: the minimum bid, the bidder’s pervious bid count, weekend
indicator, number of simultanious auctions, seller feedback, indicator for reserve, indicator for bold auction title, indicator for picture in auction. All models are OLS with Robust standard errors clustered at the zipcode level. Significance indicators: p ≤ 0.10 : ∗,
.p ≤ 0.05 : ∗∗ and p ≤ .01 : ∗ ∗ ∗.
28