Docstoc

The Effect of Free Internet Advertising on the Pricing of Used Cars Daniel A Epstein January 30 2007 This paper examines the

Document Sample
The Effect of Free Internet Advertising on the Pricing of Used Cars Daniel A Epstein January 30 2007 This paper examines the Powered By Docstoc
					The Effect of Free Internet Advertising on the Pricing of Used Cars




                          Daniel A. Epstein
                           January 30, 2007
          This paper examines the question of whether sellers of used automobiles mark up

the value of their product differently when advertising is free than when advertising is

costly.

           The advent of the internet has produced a forum that allows sellers to advertise

to a large number of people for no monetary cost. Craigslist.org, dubbed “the world’s

biggest free bulletin board,” is an extremely popular website forum that does not charge

patrons to advertise. This new method of advertising may have an influence on the

pricing of the goods advertised, and incentives may be different in the world of free

advertising. This paper intends to compare the prices and values of similar goods when

advertising is free and when advertising is costly to see if there is any difference in

markup percentage. This research is of interest as it examines the impact of the internet

on pricing policies and sheds light on the role of information’s affect on market

dynamics.

Literature:

          No research has attempted to answer whether similar goods are priced differently

dependant upon whether they have been advertised freely or at cost. There is, however,

literature on the effect of advertising on prices, on price as a signal of quality, on the role

of information gatekeepers on the internet in competitive markets, on auction theory, and

on the role of information dispersal on prices; all of which are relevant to my research.

          Benham concluded in his article, “The Effect of Advertising on the Price of

Eyeglasses,” that advertising does influence pricing, and when advertising was allowed in

the case examined, lower prices followed. He concludes that consumer’s knowledge of
market information was more important than previously thought, and that the role of

advertising cost was less important than previously thought.

       Benham’s research compared pricing in markets that allowed advertising and

markets that did not. My research compares pricing in markets that advertise over

different advertising mediums. According to George Stigler, price dispersion in a market

is evidence of ignorance of market alternatives. With perfect perception of the market

(i.e. costless information), consumers would surely buy their used car at the lowest price

available; but since buyers do not have perfect information, and since search within the

market place is usually costly and hence incomplete, sellers are able to sell their cars at

varying prices within what is a “homogenous” market. This might not be a perfect

assumption since sellers that advertise at cost often provide extra services like

warrantees, financing, and quality inspections; however, Stigler argued for this

assumption when he wrote that, “it would be metaphysical, and fruitless, to assert that all

dispersion is due to heterogeneity.”

       Stigler recognized the importance of advertising in eliminating buyer ignorance.

Laurent Linnemer and Asher Wolinsky had similar ideas when they wrote on price as a

signal of quality. They understood that pricing was dependant upon the existing

proportion of informed consumers. They also suggested that a high price can signal high

quality to otherwise uninformed consumers. This is important to remember in the used

car marketplace (which is notorious for information asymmetry between buyers and

sellers). Similarly, they saw that markups were greater as information was poorer.

       Auction theory applies to my hypothesis. The Dutch auction (also known as the

descending auction) is specifically relevant. The Dutch auction starts at a prohibitively
high price. Then the auctioneer slowly decreases the price until eventually a buyer bids.

That first bid wins and ends the auction. The Dutch auction is most commonly seen in

IPOs (and most famously with Google), but is applicable elsewhere.

Methodology:

        I considered two markets that differed in the cost of their advertising medium. I

examined the used car market as advertised in newspapers and magazines (which is

costly for the advertiser) and as advertised over craigslist.org (which is free for the

advertiser). From the car’s advertised value, I subtracted the appraised value (according

to the Kelly Blue Book). Then I took the difference and divided that by the appraised

value to determine the percent by which the value was marked up over the appraised

value. Determining a percentage rather than a level number is important, because value

may be correlated to the amount the good is marked up.

        I performed this measurement using advertised values from sources like

newspapers and magazines, where advertising is not free as well as using advertised

values from craigslist.org, where advertising is free. I then took these two groups of

values and did a two sample t-test with unequal variances to test the means.


(paid advertising value) – (appraisal)                 (free advertising value) – (appraisal)
--------------------------------------------   vs.    --------------------------------------------
             (appraisal)                                            (appraisal)


Ho: µmarkup%@cost - µmarkup%free = 0
Ha: µmarkup%@cost - µmarkup%free ≠ 0
       A simple statistical test of the null hypothesis that these groups of values are equal

answered whether the advertised values of both of these groups are marked up equally or

not.

Hypothesis:

       I expect to find that the automobiles advertised on craigslist.org are marked up

slightly more than those advertised through costly mediums. The logic behind this

expectation is the following: For every day that a seller has their ad in the paper, they

pay a fee. Thus, they have an incentive to have their ad in the paper for as short a time as

possible. One way to minimize this time is to lower the asking price on the automobile

(thus increasing the crop of potential buyers and increasing the appeal of the good). If

however, the seller incurs no monetary cost in advertising, then there is less incentive to

have a lower asking price because they lose less each day they do not sell their product.

So a seller can cast a high price into the market and wait. If there is no response, they can

lower the price marginally until they catch a buyer. This would essentially mirror a

Dutch Auction model.

       If we graphed accumulated cost over time (which includes advertising and

waiting costs), the curve of the seller that advertises freely should have a slope that is

slightly positive (only accounting for the cost of waiting). The curve of the seller that

advertises through costly mediums should have a more positive slope. If we graph the

advertised price of the good over time, we should see a slightly downwards-sloping line

(the slope is negative due to the seller’s scarce patience and the depreciating value of

their product). Now if we superimpose the curve of the advertised price onto the graph of

the accumulated cost over time, we see two converging curves. The seller wants to
maximize the gap between the advertised price (approximate revenue) and the

accumulated cost. Thus, other things being equal, the seller has an incentive to sell their

good as close to time=0 as possible. Contrarily, for the seller that advertises freely, the

only determinant of their asking price is their own patience (Graphs 1 & 2). If the seller

advertises a 1981 Honda for $384,058, the likelihood of them selling that car within their

lifetime is almost nil. So we see that the seller that advertises at no cost essentially faces

an expected value problem. They must multiply their asking price by the probability of

selling the car at that price within a desirable time frame.

               E[S|P]

               where S is the probability of selling the car within the
               seller’s desirable time frame and P is the selling price of the
               car.

If the seller has perfect perception of the market, they will be able to maximize their

profit within their desired timeframe. However, perfect perception is a false assumption.

So as I said, I expect the reality to be that free advertisers cast their marked up price into

the market and wait; lowering the price marginally until they catch a buyer. So then the

slope of the seller’s advertising price should be negative over time; determined by the

depreciating value of their good, their desired time frame for selling their good, their

patience, and their perception of the probability of selling their good at a given price

within their desired time frame.

       This model assumes that the price of advertising at cost is not negligible with

respect to the good for sale and that the cost of advertising at no cost is negligible with

respect to the good for sale. Further research might determine the magnitude to which
these assumptions are or are not true, and the extent to which they change the slope of the

selling price over time.

Results:        Table 1

Ho: µmarkup%@cost - µmarkup%free = 0
Ha: µmarkup%@cost - µmarkup%free ≠ 0

 t-Test: Two-Sample Assuming Unequal Variances

                                                   Markup                Markup
                                                percentage of         Percentage of
                                               cars advertised       cars advertised
                                                   @ cost                 freely
 Mean                                                    0.423                  0.042
 Variance                                                0.452                  0.045
 Observations                                                39                    39
 Hypothesized Mean Difference                                 0
 df                                                          45
 t Stat                                                  3.373
 P(T<=t) one-tail                                      <0.001
 t Critical one-tail                                     1.679
 P(T<=t) two-tail                                      <0.010
 t Critical two-tail                                     2.014

Discussion:

        In looking at the results, we can reject the null hypothesis that the mean markup

percentages in both groups are equal. We can reject the null hypothesis with extremely

high confidence. The resulting P-value would satisfy an α of 0.01.

        Though the mean markup percentages are not equal, the results are inconsistent

with my hypothesis. The results show that the mean markup percentage of cars

advertised at cost are almost an order of magnitude larger than the mean markup

percentage of cars advertised freely.

        Why do cars advertised at cost tend to be marked up more than those advertised

freely? What are the alternative explanations and what are the implications of these

hypotheses? There are a number of potential answers to this question. I think that a big
reason for the markup disparity is a different group of sellers using the different

advertising mediums. While there is an overlap in the mediums, I think that those who

advertise through costly mediums tend to be expert sellers, and those who advertise freely

tend to be amateur sellers. The expert sellers are well versed in the market and have

better perception of the limit to which they can markup their products. I think that the

amateur sellers have poor perception of the market, and as such do not realize how much

they can markup their product. Furthermore, those non-expert sellers may look for a

guide to find out how valuable their product is. I think that that guide is the Kelley Blue

Book (or some similar valuation resource). In other words, it is possible that the amateur

sellers, in their ignorance of the market, look to the Kelley Blue Book value to figure out

what price they will advertise their car at. If this were true, the mean markup percentage

of used cars advertised over craigslist.org would not be statistically unequal to 0. This is

exactly what we see in the statistical test in table 2.

        In running this test we see that we cannot reject the null even at α = 0.2. This

means that the mean markup percentage of cars advertised freely is statistically not

unequal to the Kelley Blue Book value. It is possible that with a greater sample set, this

would not be the case, but it seems apparent that at least a certain portion of

“craigslisters” are using the Kelley Blue Book (or some similar valuation resource) in

their pricing.

        This raises another issue. Buyers are most likely segregated, creating a selection

bias among buyers. It is possible that two different buyer populations conduct their

product search via the two different advertising mediums in question. Perhaps those who

conduct their product search through costly advertising mediums are willing to pay more
above Kelley Blue Book value than are those who conduct their product search over

craigslist.org. Clearly this must be the case; otherwise there would be no reason that cars

with high markup percentages would ever be sold. That is, unless the products advertised

over different mediums were not perfectly equal.

       The inequality of these products (or heterogeneity) is another issue. Stigler took

on this issue by saying that we cannot learn anything from our observations unless we

assume homogeneity, but in this case I think it is important to acknowledge the fact that

those who advertise at cost are oftentimes offering a slightly different product. Used cars

advertised in the newspaper or auto trader magazines oftentimes offer warranties, safety

inspections, and a certain sense of credibility in their operation. Dealers do advertise

over free forums like craigslist.org, and individuals do advertise at cost, but craigslist.org

is dominated by individual sellers, and costly advertising is dominated by dealers. I tried

to account for selection bias in products by comparing markup percentages as opposed to

advertised prices, but to fully account for extras like warranties and safety inspections,

more thorough research is needed. Xavier Gabaix and David Laibson see these

warranties and safety inspections as noise that expert sellers insert into their product to

increase ignorance, thus allowing for greater markups and greater profitability. With data

on the value of these add-ons, multi-linear regression might be able to get a more

accurate measurement of the disparity of mean markup percentages.

       Another point of note is seen when observing the actual advertisements on

craigslist.org. Many of them suggest that they are selling their product so cheap because

they need to get rid of their product quickly (due to change of location, divorce, loss of

job, etc.). I addressed this in my hypothesis when I discussed the desirable time frame of
sale. I think I underestimated the number of people who had a relatively short time

frame. In comparison, used-car dealers have a very large time frame since they have

storage space and are operating in scale. Changing the time frame assumptions changes

the whole dynamic of my hypothesis and pricing in general.

       Craigslist.org’s extremely easy access may also be a factor in significantly more

competitive pricing. The sheer number of different sellers that advertise on craigslist.org

(many of whom are not solely interested in maximizing profit) can drive down prices.

Newspapers and magazines may offer more total cars, but many of those cars are

advertised by the same seller. Craigslist.org is packed with individual non-colluding

sellers, all competing with each other.

       The internet is a goldmine of market information. Consumers can gain a much

more accurate perception of the market with an equal level of effort by using the internet

than by searching newspapers and magazines. This assertion matches “expert buyers”

with amateur sellers and “amateur buyers” with expert sellers. It is no wonder then that

we see the mean markup percentages diverging in the direction that we do. It should be

noted though that some newspapers do post their “Classifieds” section on the internet,

which may bring into question the power of this assertion.

       Negotiation expectations should also be considered. Unlike many goods, car

prices are starting points of negotiation. The realities of negotiation impart some

complex dynamics with respect to pricing; especially considering the fact that many cars

bought from dealers are sold as a part of a deal that includes a trade-in.
       These alternative hypotheses were not directly tested, but the magnitude of the

difference in mean markup percentage between these two groups appears to be an

anomaly.

Conclusion:

       My initial hypothesis was that pricing for used cars advertised freely over the

internet would operate like a Dutch auction, resulting in high initial offers. I looked at 39

cars advertised freely over the internet and 39 cars advertised through costly newspapers

and magazines and found the mean markup percentages to be 4.2% and 42.3%

respectively; a disparity of 38.1%. This result proved wrong my initial hypothesis.

Alternative hypotheses concerning segregation among sellers and buyers, product

heterogeneity, the role of information, seller time horizons, barriers to sale, and

negotiation peculiarities were considered.
                                     Bibliography

Athey, Susan and Philip Haile. “Identification and Standard Auction Models.”
       Econometrica Vol. 70, No. 6 (November, 2002), 2107-2140. 18 Nov. 2006
       http://www.jstor.org/view/00129682/sp030003/03x0041o/0?frame=frame&userI
       D=80fcb6a5@wustl.edu/01cc99332100501b1a752&dpi=3&config=jstor.

Baye, Michael R. and John Morgan. “Information Gatekeepers on the Internet and the
      Competitiveness of Homogeneous Product Markets.” American Economic
      Review Vol. 91, Issue 3 (June, 2001), 454-474. 18 Nov. 2006
      http://ideas.repec.org/a/aea/aecrev/v91y2001i3p454-474.html.

Benham, Lee. “The Effect of Advertising on the Price of Eyeglasses.” Journal of Law
      and Economics 1972, 337-352. 18 Nov. 2006
      http://www.jstor.org/view/00222186/ap020022/02a00040/0?frame=frame&userI
      D=80fcb6a5@wustl.edu/01cc99332100501b1a752&dpi=3&config=jstor.

Gabaix, Xavier and David Laibson. “Some Industrial Organization with Boundedly
      Rational Consumers.” Working paper, Current Draft: December 27, 2003, 1-74.
      18 Nov. 2006
      http://post.economics.harvard.edu/faculty/laibson/papers/briodecember30.pdf.

Jones, Tamara. “It might be the world’s biggest ‘lost and found’.” The Hamilton
       Spectator November 8, 2006. 19 Nov. 2006
       http://www.hamiltonspectator.com/NASApp/cs/ContentServer?pagename=hamilt
       on/Layout/Article_Type1&c=Article&cid=1162939816396&call_pageid=102042
       0665036&col=1112101662670.

Kaufmann, Lutz and Craig R. Carter. “Deciding on the Mode of Negotiation: To
     Auction or Not to Auction Electronically.” The Journal of Supply Chain
     Management Spring 2004, 15-26. 18 Nov. 2006 http://www.blackwell-
     synergy.com/doi/pdf/10.1111/j.1745-493X.2004.tb00166.x?cookieSet=1.

Linnemer, Laurent. “Price and advertising as signals of quality when some consumers
      are informed.” International Journal of Industrial Organization Vol. 20, Issue 7
      (September 2002), 931-947. 18 Nov. 2006
      http://www.sciencedirect.com/science/article/B6V8P-45PK0T2-
      2/2/a4e9de9cf7b58aedc2eda2248c82ff5b.

Shiller, Akerloff, Choi, Laibson, Madrion, Metrick, Alesina, Angeletos, Mankiw, Reis,
        Wolfers, Gabaix, Chirinko, Schaller, Di Tella, and MacCulloch.
        “Macroeconomics and Individual Decisionmaking.” NBER Reporter Winter
        2003/2004. 18 Nov. 2006 http://www.nber.org/reporter/winter04/winter04.pdf.
Stigler, George J. “The Economics of Information.” The Journal of Political Economy
        Vol. LXIX, No. 3 (June, 1961), 213-225. 18 Nov. 2006
        http://www.jstor.org/view/00223808/di950876/95p0002d/0?frame=frame&userID
        =80fcb6a5@wustl.edu/01cc99331400501b27fa7&dpi=3&config=jstor.

Wolinsky, Asher. “Prices as Signals of Product Quality.” Review of Economic Studies
      (1983) L, 647-658. 18 Nov. 2006
      http://www.jstor.org/view/00346527/di990666/99p0238w/0.
Appendix:

Graphs 1 & 2
Table 1:

Ho: µmarkup%@cost - µmarkup%free = 0
Ha: µmarkup%@cost - µmarkup%free ≠ 0

 t-Test: Two-Sample Assuming Unequal Variances

                                             Markup             Markup
                                          percentage of      Percentage of
                                         cars advertised    cars advertised
                                             @ cost              freely
 Mean                                              0.423               0.042
 Variance                                          0.452               0.045
 Observations                                          39                 39
 Hypothesized Mean Difference                           0
 df                                                    45
 t Stat                                            3.373
 P(T<=t) one-tail                                <0.001
 t Critical one-tail                               1.679
 P(T<=t) two-tail                                <0.002
 t Critical two-tail                               2.014



Table 2:

Ho: µmarkup%free = 0
Ha: µmarkup%free ≠ 0

 t-Test: Two-Sample Assuming Unequal Variances

                                          Markup
                                       Percentage of
                                            Cars                   0
                                        Advertised
                                           Freely
 Mean                                           0.042                           0
 Variance                                       0.045                           0
 Observations                                      39                          39
 Hypothesized Mean Difference                       0
 df                                                38
 t Stat                                         1.250
 P(T<=t) one-tail                               0.110
 t Critical one-tail                            1.686
 P(T<=t) two-tail                               0.219
 t Critical two-tail                            2.024
Tables 3 & 4:

  Descriptive Statistics of price when advertised at
                          cost

Mean                                       8939.051
Standard Error                             1078.629
Median                                          6977
Standard Deviation                         6736.034
Sample Variance                         45374152.37
Kurtosis                                      -0.434
Skewness                                       0.768
Range                                         23140
Minimum                                         1350
Maximum                                       24490
Sum                                          348623
Count                                             39
Confidence Level(95.0%)                    2183.570

   Descriptive Statistics of price when advertised
                        freely

Mean                                       8919.513
Standard Error                             1172.387
Median                                         6000
Standard Deviation                         7321.552
Sample Variance                         53605126.62
Kurtosis                                      1.335
Skewness                                      1.221
Range                                         32000
Minimum                                         500
Maximum                                       32500
Sum                                          347861
Count                                            39
Confidence Level(95.0%)                    2373.373
Tables 5 & 6:

  Descriptive Statistics of Kelley Blue Book value of
               cars advertised at cost

Mean                                       7247.821
Standard Error                              992.896
Median                                         5420
Standard Deviation                         6200.636
Sample Variance                          38447889.2
Kurtosis                                      0.618
Skewness                                      1.155
Range                                         22320
Minimum                                         525
Maximum                                       22845
Sum                                          282665
Count                                            39
Confidence Level(95.0%)                    2010.014

  Descriptive Statistics of Kelley Blue Book value of
                cars advertised freely

Mean                                       8416.205
Standard Error                             1094.188
Median                                         6040
Standard Deviation                         6833.200
Sample Variance                         46692628.22
Kurtosis                                      1.482
Skewness                                      1.348
Range                                         28760
Minimum                                        1030
Maximum                                       29790
Sum                                          328232
Count                                            39
Confidence Level(95.0%)                    2215.067
Tables 7 & 8:

 Descriptive Statistics of markup percentage of cars
                  advertised at cost

Mean                                          0.423
Standard Error                                0.108
Median                                        0.264
Standard Deviation                            0.672
Sample Variance                               0.452
Kurtosis                                     10.590
Skewness                                      2.942
Range                                         3.801
Minimum                                      -0.475
Maximum                                       3.326
Sum                                          16.507
Count                                            39
Confidence Level(95.0%)                       0.218

  Descriptive Statistics of markup percentage of
              cars advertised freely

Mean                                        0.042
Standard Error                              0.034
Median                                      0.041
Standard Deviation                          0.212
Sample Variance                             0.045
Kurtosis                                    0.474
Skewness                                   -0.188
Range                                       0.985
Minimum                                    -0.515
Maximum                                     0.470
Sum                                         1.655
Count                                          39
Confidence Level(95.0%)                     0.069