It’s Just Like Telling a Joke

From the O*Net database, i learned something today — oral skills presentation is the most important attribute in an economist is oral communication. We joked about this in seminar today, but in a sense its true. Once you’ve written a paper, you must explain it to people. 

This reminds me of a joke told by Conan O’Brien when he was on Leno’s show. Conan shares something he learned from Stan Laurel (Laurel and Hardy) about telling a joke: ”Always do this. Tell the audience what you’re going to do. Do it. And then tell them it has been done.”

Tell them what you are going to do! How is the paper related to other papers? Lots of people ask this question. You want to reassure people that your work is part of the larger story of scientific advance. This is also a practical thing as well: when you want to come up with new work, you look at whats been done and think of ways to extend them.

 Do it! What is the motivation? Really good papers are of the small model kind — present stylized facts and construct a model as a way to interpret the data, and then test your interpretation. Its good to have some kind of empirical regularity at the heart of your paper. There are exceptions, such as pure theory or pure empirical/econometric work.

Tell the its been done! What is the intuitive interpretation of your model? Relate your model to the motivation you were talking about earlier, and to the results of the papers that inspired you that you copied from.

Criminal Copyright?

Great article on copyright and fairuse. But, this paragraphed called out to me:

This war must end. It is time we recognize that we can’t kill this creativity. We can only criminalize it. We can’t stop our kids from using these tools to create, or make them passive. We can only drive it underground, or make them “pirates.” And the question we as a society must focus on is whether this is any good. Our kids live in an age of prohibition, where more and more of what seems to them to be ordinary behavior is against the law. They recognize it as against the law. They see themselves as “criminals.” They begin to get used to the idea.

That recognition is corrosive. It is corrupting of the very idea of the rule of law. And when we reckon the cost of this corruption, any losses of the content industry pale in comparison.

I agree. Its not the big things, but the small things, where the temptation to ‘rationalize’ rule breaking might happen in a culture of excessive criminalization of copyright violations

constraints

Every economics student learns the lagrange method of solving constrained optimization problems. Beyond the technicalities, the lesson is clear, if not obvious — When u are constrained, it stops you from making choices you’d made if you were unconstrained. Hence a constraint constrains you, and the math help figure out by how much. I was just reading a textbook description of a model that prohibits the return on capital to rise above a certain level, say s. There are two, and only two, possibilities: either it doesn’t matter because s is too high to constrain, or it matters greatly, and the capital decision is made to meet this constraint.

UPCs

Broda and Weinstein’s new paper on international price differences tells us the source of country price dispersion between US and Canada

They study this via UPCs obtained from AC Neilsen. The first key result is that UPCs sold in BOTH the US cities and Canadian regions follow the LOOP relatively well. The other key result is that common UPCs is quite small. For the US city pairs, the highest pair with w NY-Phili, with only 30% in common. The further the city pairs are to each other, the lower the fraction of common UPCs are: about 15% for LA-NY.

For US-Canada pairs, the slope of the distance-common UPCs line is flat — no matter what the distance is, the range is from 5% to almost 10%.

For Canada pairs, the range is wider (from 33% to 55% common UPCs), and the minimum is much higher than the maximum of the US city pairs.

These results are consistent with higher price variation within the US vs Canada. Broda/Weinstein say that the fact that common UPCs are smaller for cross-border pairs is the key reason why we observe deviations from LOOP in aggregated prices cross border. This is the result of Tesar and Gorodnichenko also, although they don’t know why the variation would be different.

Broda-Weinstein move this debate one step forward — now we know why Canada deviation is different from US variation in prices. Its about what UPCs are sold where. The next question is — why is product variety so different? One guesses that this is about retailers and manufacturers price discriminating via changing the goods characteristics. Examples: eggs can be large, medium and small, come in several pack sizes, free range or not, etc. BW report that fresh eggs have 2,275 possible varieties, while the ‘typical’ group has about 2,700 varieties.

Also, among US cities, the composition of goods is different, and is more different the further away, the more different it is. For Boston-LA, about 2,700 miles, the common UPCs are only 17%, while for Boston-NY, distance of 230 miles, the common UPCs are 24%. What is the source of this non-integration/State-City Bias?

Optimal Stopping

When do you know when to quit? When to buy? This is a hard question, so as first pass, we make some assumptions.

Money has value R that you discount at rate r. The cost of the good is C, and the cost falls at rate (r+w). This is a process innovation that makes producing the good cheaper over time, and, we assume that this savings is passed on to consumers. Clearly, there is a benefit to waiting because cost is falling to zero. A standard method for discounting is exponential discounting, which gives the result that cost is falling faster earlier in time, and slower later in time. Last, the good itself has value S, which the person discounts at rate r, but enjoys from time T to forever (we assume infinitely durable goods for convenience — it won’t affect the conclusions later).

The problem is to decide the value of T, i.e. when to exchange money for the good.

Note that it must be true that S>R, otherwise, the value of the good will forever be less than the value of money. You end up never buying, which means T is infinite.

What is the basic trade-off? At the best T, it must be that waiting longer isn’t worth it. In other words, the benefit of waiting (lower cost) is less than the cost of waiting (which is foregone benefits of owning the good, which is a function of S-R). It must also be true that buying sooner won’t help, coz the cost of buying early (high cost) is far greater than the benefit of owning the good.

So, suppose we decide the value of the best/optimal T according to the rules above. What then is is the effect of a faster rate of cost reduction (a higher w)? This should push you to wait; BUT only when the cost reduction benefits of waiting are huge, which as we’ve determined is when the optimal is small. The process improvement forces you to buy earlier if the optimal T is large.

What helps determine whether optimal T is large or small? The size of (S-R). The larger S-R is, the larger the benefit of buying sooner, coz the gain you get from buying, S, is larger than the value of money, R. The larger S-R is , the lower the optimal T — in this situation, a cost improvement leads to waiting longer.

Finally, lets say there is no secular price declines for this good, w is zero. What then? Pretend you are at time zero, deciding whether to buy. If you delay, you earn a net of R-S+rC, or R-(S-rC). If R>S-rC, you think, if i delay, i get R (money happiness), which is greater than S (good happiness) minus rC (one time cost of buying the good). The opposite holds when R<S-rC.

Evidence for the Colbert “Bump”

A paper by a UC San Diego professor on the effect of an appearance of politicians on the show’s segment “Better Know a District” on their fund raising. The show gives a significant bump to Democratic congressmen who appear on the show — they earn 37% more than similar democratic congressmen who didn’t appear on the show. The analysis is merely suggestive of a causal link. The paper is written in ‘colbert style’ and is quite entertaining to read, and is a great resource about the colbert and its “disproportionate influence” in politics.

Another blog by juice analytics studied colbert’s influence on book sales. their analysis indicates that on-average, books sales increases by 10%, and that this effect comes from the increase in sales of pop intellectual book and liberal book authors that guest on his show. I’m pretty sure that if they study “The Daily Show”, they would get a TDS bump result as well.

I wonder how powerful this effect is vis-a-vis the authors’ other opportunities to market their book? From a professor in my school, i know that an appearance in either show was heavily encouraged by his agent.

Here is a video from AP about the same phenomenon and some choice words from the man himself.

Optimal product launch strategy

There is a paper in our seminar schedule about the monopolist’s product launch strategy in multiple markets. Its highly stylized, but heck, its a start.There are N > 1 markets in which the monopolist can launch her product. The monopolists decision is to decide how many markets in which to launch. She can launch in all, in some, or none of these markets. For some reason, (unexplained, but i assume for simplicity), price p is fixed, at say 1/2. The good is either high quality, and its value v is 1, or low quality, where in its v is 0. The monopolist doesn’t know the quality of the good, which explains why we have non-discriminatory pricing. If the monopolist knew about her products quality, she would set two different prices and there would be a separating equilibrium. [Which raises an interesting question: how can a monopolist not know the quality of the product?]

The consumers belief about the value (lambda) takes values from (0,1). The consumer buys v-p>0 [This is confusing, it should be lambda-p>0? Is lambda the distribution of consumer's unit demands? I think the author wants to say that lambda is the same for each consumer in a market, which means its just like assuming one consumer per market, not terribly interesting].

Whats interesting is the updating method. Imagine you are in a market, and you observe a ’signal’ (she doesn’t explain what this signal might be, but this is a random signal) from another market where the product is launched. This signal can either be good or bad. A good signal may mean that the value of the good is 1 (the highest possible). A bad signal is defined similarly. A signal is independently and identically distributed across markets and time.

However, like many signals in life, signals are noisy. Let q = Prob(signal=1|v=1) =Prob(signal=0|v=0). That is, q is the probability that the signal is correct.We assume here that signals are more likely to be correct than not: q>1-q. [note: if this isn't true, then the graph i present below will switch].

The updating of beliefs on the product’s value is of interest to the monopolist. The monopolist would prefer to launch sequentially, IF she can manipulate willingness to pay in markets where she has not launched yet. In terms of the signal, if the signal is good, this implies, by Bayesian updating, those who see this signal would revise their beliefs/lambdas upward which raises demand.

Put in simply, this is kinda like viral marketing. Launch a product for a small group first. They tell the world that the product is good once they consume it and know its good. The rest of the world sees the signal and says, hey, windows vista is better than i thought… so they revise their valuations upward based on this signal. Similarly, if the signal is bad, they revise their values downward.

A consumer in a market where the good has not been launched has two pieces of information. The performance in the first market(s) in which it has been launched, as well as the random signal defined earlier.

Below you find a graph depicting the bayesian updating of beliefs:

Given that the price is 1/2, if lambda=lambdalow or less, then the succeeding markets will never buy the good, no matter what the realization of the signal. Hence you will never launch the good. We say that the consumer here is extremely pessimistic.

If lambdalow<lambda<p=1/2, then something interesting happens. Here, we say the consumer is pessimistic — the prior belief is that the good is ‘overpriced’.

Assume for simplicity two markets, and we consider a sequential launch which simply means launching one market at a time. Lets say the good was succesful (which means it received a good signal) in the first market; this leads to an updating upwards for the second market. But there is also a random signal A good random signal leads people to buy, a bad random signal lowers their lambda further. So,if the signal in market one is good and the signal is market two is good, there is no difference between sequential and simultaneous launch. Both markets buy anyway. If the signal in market one is good, and the random private signal in market 2 is bad, then the two signals cancel each other out, leading to lower total sales (good sales in market 1, no sales in market 2) — an inferior outcome which is came from the fact that the product was launched sequntially. If the signal if the first market is bad, and the second is bad, no one buys — but this would be true in a simultaneous launch too.

So the dominant strategy is to simultaneously launch.

[Note: its important here that you have representative, identical consumers in each market to get this result. If you abandon these, the analysis becomes more involved, but still follows the general line of reasoning outlined above]

If p=1/2>lambda<lambda high, something interesting happens. This time, the consumer is optimistic. Here, if you have simultaneous launch, each market will receive a random signal of s, which causes some to buy, others not.

If the launch is sequential, there are two signals for the later market: success/failure in the first and a private random signal in the second market.

The reasoning is similar to the preceeding case. If the signals are opposing, then they cancel out — but thats fine because the lambda>p and they (the second market) still buys. If the signals are the same, clearly both markets buy the good. We see why sequential benefits the monopolist — its a form of random signal hedging. A sequential launch allows opposing signals to cancel outwhich leads to higher total sales, while in a simultaneous launch, opposing signals mean random successes which is inferior.

Finally, if lambda>lambdahigh>1/2, then no matter what the realization of the signal, each market buys the product. so sequential is the same as simultaneous.

This is nice model that allows us to see how sequential launches help a monopolist when there is information assymmetry both for the seller and the buyer. They way the author explains the mechanism and demand was confusing at first. Also, the next step would be to endogenize the signal as a function of market variables, like sales and quantity and price. I also don’t like that latter markets have two signals. I’d rather model it as one signal, and then each market is heterogenous along a certain dimension. In this way, we might be able to get the result that if a sequential launch is implemented, then the first market must have the larger market size, or higher profit potential. This is much more realistic.

Repugnant Transactions

Recently i saw Prof. Alvin Roth’s lecture on designing markets. We know that markets have a potential to improve welfare through a more efficient allocation of goods. One day, when economists really know what they are doing, we can design markets the way engineers can design products.

At any rate, Professor Roth is an engaging and inspiring speaker. He makes market design seem easy — which we all know isn’t the case at all. One of the things that surprised him is that people seem reluctant to engage in certain type of transactions. Some of these transactions are ‘repugnant’ — and as such are constraints to be faced by any market designer. Watch the video above to see some an overview of his work on many markets, most important of which is his work on Kidney Exchanges.

Related to this is an essay in the NYTimes about surrogate mothers in India.  To many of us, this market is disturbing; but to many Americans and Indians, this is just another example of the gains from Trade.

I once heard an anthropologist talk about repugnant markets. And the lesson i took home from this is that we (as people) create borders around our person, and everything beyond this border is potentially tradeable for money, while everything within it requires the notion of a ‘gift’. A ‘gift’ is something that is a part of you that you voluntarily share with another. ‘A part of you’ is the key idea– the gift can never be separated from the person giving and the relationship between the two parties. Meanwhile, everything else is conceptually not a part of you, and thus can be commodified. This barrier moves in time, and changes across generations, and now, even within a generation.

Price Dispersion

Price dispersion is when different sellers offer different prices for the same good in a given market. Its different from third-degree price discrimination, where a seller offers a good for different sellers for different groups of buyers. It is also different from geographical price dispersion, as in the failure of the Law of One Price across cities and countries.

I imagine a good in a city (say, soap), that sells for different prices in different stores at a given point in time. The theoretical approach is that there is imperfect information among consumers about where to find the lowest price good.

An early attempt at this is by Diamond (1977, “A Model of Price Adjustment” JET). Given that every consumer is the same [preferences] and each buyer knows only the price in one shop, and there as many shops as there are buyers. Everyone has a cost to learn about prices in other stores sequentially (imagine calling one new store, one store at a time, a process called sequential search). The Nash Equilibrium is no price dispersion, and each store charges a monopoly price. At this point, it becomes obvious. If everyone is the same, each has the same reservation price, say r, the highest price at which a buyer buys the product. Each store knows this and sets p=r. Each buyer knows this, and thus know that is no price dispersion, and does not bother to learn other store’s prices.

It is clear from the above that to achieve NE where there is dispersion, you have to exploit not only imperfect information, but also some kind of heterogeneity in seller and buyers, or both. An example is Varian (1980, A model of sales, AER). He assumes there are informed consumers and uninformed consumers, the relative proportions in the population are exogenous. The result is a mixed strategy in prices. Stores randomize because there is no dominant strategy — if all stores sold at the same price, it would be profitable to undercut your competiton and attract the informed buyers. The presence of uninformed buyers pushes stores to sell at a high price.

The next step is to endogenize the decision for consumers to inform themselves or other prices. Varian’s model can be interpreted as two groups of consumers, one with high search costs and one with low search costs. A paper by Burdett and Judd (1983, Equilibrium Price Dispersion, Econometrica) argues that even  with all consumers with a small, positive search cost, price dispersion can still exist.

Where is the Love?

This article from Time wants to explain the disappearance of the love story. There are demand side issues (We’re sick of love stories!) and supply side innovation issues (I can’t make original love stories anymore!). It concludes with a comment on the structure of the industry. Titanic, the most successful movie in movie history, was a slow starter and needed time to earn its billion dollar take. Because of competition, supply side costs, its more important than ever for films to make alot of money on its first weekend. This means you have to attract people who watch movies on their opening weekend — young men.