Friday, January 30, 2015

Dinner with Haavelmo

The student magazine at TSE (TSEconomist) asked some of us to tell them with which thinker of the past we would like to have dinner. Here is my answer.

If I was to have dinner with an economist of the past, it would probably be Trygve Haavelmo. His 1944 Econometrica paper “TheProbability Approach in Econometrics” is for me a milestone for econometrics and applied economics, and, let’s be bold, economics in general. I read it several times and continue reading it regularly. Each time, I discover a novel gem. This paper laid out the ground for applied work in economics with a sound statistical basis but also taking into account the specificities of our field, as for example the existence of interdependence and equilibrium relationships. This paper inspired the subsequent work of the Cowles Commission that forms the basis of the econometrics curriculum that we teach. 

But this paper is also so much more than theoretical econometrics. It is a paper about epistemology, how and why we can do research in economics. I think the views Haavelmo expressed in this paper are so deep that they implicitly structured how applied research in economics has been conducted until the late 80s. And in some of his views, you also see the nascent credibility revolution coming: Haavelmo coined the term natural experiments (actually, Josh Angrist, the modern proponent of this approach, directly attributes the term to him) and he also insisted heavily on setting up a design of experiments with each economic theory, because, in his own words: “Our guard against futile speculations is the requirement that the results of our theoretical considerations are, ultimately, to be compared with some real phenomena.” He also said that “What makes a piece of mathematical economics not only mathematics but also economics is, I believe, this: When we set up a system of theoretical relationships and use economic names for the otherwise purely theoretical variables involved, we have in mind some actual experiment, or some design of an experiment, which we could at least imagine arranging, in order to measure those quantities in real economic life that we think might obey the laws imposed on their theoretical namesakes”. 

What I would love to talk to him about is why he did not promote much more the experimental method in economics. My understanding of his paper is that, at some point, he abandons all hope of ever conducting experiments in economics and settles for something else, namely modelling what we do not know, the confounding factors. He uses an analogy with physics, saying that we have to model the forces of friction because we cannot run experiments in a vacuum. I think this approach was wrongheaded and contributed to mixing science with engineering for a long time in economics. I would also ask him why, despite the repeated failures of the approach that he proposed, and that he acknowledged repeatedly, in his 1958 speech  as a president of the EconometricSociety, and in his 1989 Nobel prize speech, he blamed the theory instead of blaming the method and putting experiments back into the picture. I would also love to discuss with him the exciting direction that economics is taking now, with more use of experiments for testing theories and policies, and the novel and welcome approach of testing the predictions of structural models againstpolicy changes.

Finally, because Trygve did not seem like a funny guy, I might also invite David Hume to take part. Apart from being my favorite philosopher, he was nice and charming fellow and would for sure cheer us up. Oh, and he also basically invented the notion of counterfactual that is crucial for causal analysis. That would be some dinner!

Monday, January 26, 2015

Land reallocation and the structural transformation of the French agriculture

I have recently started to work on an exciting new project about land reallocation and the structural transformation of French agriculture. In this post, I'll present the motivation for the project and how I plan to conduct  it.


Understanding why countries are poor or rich, and how to unleash their growth opportunities is a key question in economics, for obvious reasons. One theory explaining why some countries are poor is factor misallocation. For some reason, factors as land or capital may end up being distributed unefficiently - maybe divided in too small farms or too scattered pieces, or not allocating enough land or capital to the more able managers. According to this theory, reallocating factors in a more efficient manner would unleash strong growth. There is growing evidence that  some kind of factor misallocation is at play, but we lack direct evidence that a policy of active reallocation of factors of production can indeed increase productivity. This is key because providing evidence of factor misallocation does not mean that a policy of reallocation would increase productivity. Moreover, studying an actual event of land reallocation might help in undertsanding how factor misallocation impairs productivity growth. Is that because of poor organization of factors of production, as for examples farms divided in scattered pieces, or because very productive farmers were not allocated enough land, or because it unleashed the realization of more productive investments? Finally, such an approach might also shed light on the sources of misallocation: why the market forces were not enough to allocate the factors of production efficiently?

To shed light on this issue, my project is to use a major land reallocation experiment that took place in France, called "reparcelling" (remembrement) . Between 1944 and 2006, roughly 18 million hectares of agricultural land - more than half of the French Usable Agricultural Area -  have been reallocated in France. This is, I believe, the major land reallocation event in France since the French Revolution. The basic principles of land reallocation have been edicted by the Vichy government in March 1941. The main aim was to decrease morcelling by regrouping land around each farm in order to decrease the transportation time between plots of land and to have larger and more regular plots of land more suitable for the adoption of modern technologies as the tractor and the plough. Concretely, the process of land reallocation takes place at the commune level, the smallest administrative unit in France, roughly of the size of a US census tract. Each plot of agricultural land is attributed to a class reflecting is quality. The project of land reallocation is proposed by a professional surveyor and has to satisfy an equivalence constraint: the area of each class of land that each farmer possesses has to be the same in both the initial and the final allocation. After reallocation, each farmer thus possesses the same amount of land of the same quality as before, but regrouped in larger plots around his house.

The first key question that this project seeks to answer is whether land reallocation has caused a surge in agricultural productivity. At first sight, the evidence seems extremely favorable. Over the same period as that of land reallocation, French agriculture has undergone a deep structural transformation. Between 1950 and 1997, the number of tractors has increased ten-fold, the average farm size has almost tripled and yields have skyrocketed (multiplied by 6 for maize, by 4 for wheat). At the same time, the share of the active population working in the farming sector decreased from 27% to 4% and the number of farmers decreased from 2.3 million to .7. This phenomenon has been coined "The End of the Paesants" by the French sociologist Henri Mendras and "The Quiet Revolution", by the leader of farmers' unions Michel Debatisse.

Without further investigation, though, it would be misleading to attribute the bulk of the structural transformation of French agriculture to land reallocation. Indeed, many other changes might explain part of the exceptional structural transformation. First, other agricultural policies where put in place at the same time as land reallocation was encouraged. The very important Laws of Agricultural Orientation of 1960 and 1962 due to the General de Gaulle and his Ministry of Agriculture Edgar Pisani have not only facilitated land reallocation, they have created a host of other policies aiming at increasing farm size and spurring investment in agriculture. A first set of policies altered the land market, favoring renters over owners. Starting in 1945, land rents were supervised with limited price increases, investments could be decided independently by the tenant which had to be compensated in case the investment was not fully amortized at the end of the lease agreement, and the tenant had the right to preempt the land he rented in case the owner wanted to sell. In 1970, longer lease terms of 18 and 25 years were introduced. A public structure (the SAFER) could preempt any land transaction and sell its holdins to a candidate of its choice. A second set of policies favored the installation of younger farmers on farms of larger sizes. A minimum installation size (MIS) was defined and the SAFER had to sell their holding favourably to farms larger than the MIS. Starting in 1962, a set of subsidies for early retirement was put into place, with the aim of freeing land occupied by small farmers for young farmers. Subsidized loans were also directed towards farmers with the larger farm sizes. Starting in 1973, a subsidy for the installation of young farmers is created. In 1962, subsidized prices were introduced as part of the Common Agricultural Policy. Over this period, a strong movement of adaptation of (mainly U.S.) agricultural innovations was set into motion with the creation of INRA in 1946 and of the CNEEMA in 1955. These innovations were then transferred to farmers by a very dense network of farmers' groups, the CETA.  (And yes, the French love acronyms.)

In order to separate the effect of land reallocation from that of other concomitant policies, I plan to use the slow progression of land reallocation over time and space. By comparing the evolution of agricultural productivity of similar communes over time, some undergoing reparcelling and some not, or at a later date, I can separate the effect of land reallocation from that of other concomitant policies, since these other policies should affect these similar communes similarly.

An interesting feature of reparcelling is that it has affected land allocation in two ways. In the short run, it has reallocated land keeping total hodlings and quality constant. The only effect has been to regroup land into larger plots closer to the farm headquarters. The first effect of reparcelling is thus to correct for misallocation of land across space. The second effect has been (I suspect) to increase farm size by facilitating land transactions between old and young farmers. The first type of reallocation effect has never been documented to my knowledge. It would be interesting to disentangle both, but in a first step, I only plan to study the combined effect of both these types of reallocation.

Next steps

The first important step is to build a database of all the land reallocation operations in France since 1945 at the commune level, with information at least on the date of the operation. This has been kindly provided by Nadine Polombo, who has worked extensively on this topic with her colleague Marc-André Philippe. They have compiled and updated a database maintained by the French Ministry of Agriculture.

The next step for me is now to study in detail the progression of land reallocation over the years and over space and to relate it to several characteristics of farms. One easy thing to do is to look at the differences in farm structure in 2000 between communes that have been reparcelled and commune  that have not. One problem is that these communes may differ for reasons other than reparcelling. One important fact about reparcelling is that it is mainly confined to the northern part of France. Several possible explanations have been given for this fact. For example, land reallocation operations took place preferentially in open flat land of homogeneous quality. It also seems that regions with a long tradition of land rental witnessed more reparcelling and started it earlier. As a result, communes that have reparcelled were initially different: flat land favorable to crops and high yield varieties, maybe already larger farms. Even in the absence of reparcelling, these communes would have evolved differently. We say that these differences confound the effect of reparcelling. 

In order to minimize the impact of the confounding differences, I plan to use strategies:
1/ Compare communes located closely enough to each other, in areas with similar productive characteristics,
2/ Compare the evolution over time between communes that have reparcelled and communes that have not. This approach is called "difference in differences," since we compare how differences between communes change over time. Difference in difference works when the effect of confounding factors remains fixed over time. This might not be an attractive assumption since it might very well be the case that structural change would have been quicker and steeper in reparcelling communes. I will have to explore other strategies for modeling the evolution of modernization in the absence of reparcelling. There are several econometric techniques around which might be suitable. One key condition for applying this method is to access data on older farm censuses. Accessing the first farm census of 1955 would be ideal, since the bulk of land reallocation took place after that date and we thus have an initial benchmark to which to compare the subsequent evolution.
3/ Compare communes that have been reparcelled to communes that have not for reasons uncorrelated with the modernization of agriculture. For example, around 10% of all reparcelling operations have taken place along major infrastructure projects as roads or railroads. It is possible to estimate the effect of reparceloing by comparing communes located close to infrastructure projects to communes located farther away.


Echoes from the European Summer Syposium in Labor Economics

I have not posted much recently. I am back on the tracks now. Here is a post I could have published a long time ago relating the papers that have been presented at a symposium I attended this fall.

I have spent three days at the ESSLE. This has been a really nice event extremely well organized in a delicious place. I've met really nice people, received encouragements and comments on my own research and have been exposed to some nice work. Let me here give a brief overview of some of the papers and posters that I have been exposed to. As I was presenting my own work in a poster session, I was not able to attend some of the presentations. Also, some of the work presented was too far from my own area of expertise for me to be able to comment on it.

Christina Gathmann showed that mass layoffs affect negatively local areas not only through the already documented negative long term effects on laid off workers, but also decrease total local employment in other firms. Javier Ortega brought evidence that immigration seems to have detrimental effects on the wages of blue collar workers in France. This would suggest that all the protests against immigration that we hear on the extreme right might have some empirical support. The effect seems to be especially major for construction workers. Christian Fons-Rosen presented evidence that former employees of big firms explore more risky and new ideas when they start their own firms, and produce higher quality research but with more variance. It seems to suggest that big firms are not an engine for radical ideas. Thomas Le Barbanchon showed that hiring subsidies (exemption of social taxes) have boosted employment in French firms during the recession, especially in slack labor markets. This is important because it shows that supply side reforms might have an effect even when demand is depressed. Ankita Patnaik showed that the move to a compulsory parental leave of 5 weeks for daddies in Quebec (required minimum amount of daddy's leave duration for mums to be able to take their own leave) generated not only an increase in the duration of daddy's leave by 3 weeks from an initial level of 2 weeks but also seems to have generated strong changes in intra household behavior that lasts after the parental leave period is over: dads spend less time at work, take more time for chores and earn less money whereas mums earn more money, do less chores and spend more time with their families. In my opinion,the results on earnings were very large, suggesting that there might be selection bias due to couples delaying or accelerating the decision to have a child because of the reform. Andrea Bassanini showed that in France the probability of layoffs seems to increase with the distance of a plant to its headquarters. This might be because of local social pressure not to hurt local employment or as a service to the local authorities. Kristina Huttunen showed evidence that offshoring might have negative effects on employment across industries but seem to have a positive effect within industries: firms that offshore part of their production seem to expand. The reverse is true for import competition. Nishith Prakash shows that you can obtain a substantial increase of enrollment in high school by girls in India by buying them a bike to ride to school. 

The second day, Stefania Albanesi documented that the increase in women labor force participation is slowing down in the US and in other OECD countries while at the same time the decrease in the gender wage gap slows down. She related that to the increase in top wages: women married to men whose wages have rapidly increased in the last decades do not need to work and thus take more time off the job market, yielding to lower salaries. My attention somewhat decreased thereafter, not doubt because the time of my own presentation was getting closer but also because I am less familiar with the topics covered. I apologize to Pietro Garibaldi and Gerard Pfann for not reporting on their work. Konstantinos Tatsiramos showed that the effects of school and community of people lifetime earnings seemed to stem mainly from the ignorance of the fact that family self select to schools and neighborhoods. Once taking this correlation into account, he finds that school and community do not seem to have strong effects on lifetime earnings. Ulf Zolitz presented in the poster session a very nice paper using the random assignement of students to classes to show that low achieving students benefit from having better classmates, but not the best classmates. The negative effect of the best students on the weak students is an interesting finding. Actually, Ulf has also a paper on the negative learning effects of cannabis usage. He cleverly exploits a local law forbidding French citizens to consume pot in the Netherlands. 

Finally, the third day, Erich Battistin presented a paper revisiting the famous Maimonides' rule approach for estimating the effect of class size on students' achievement. I am not going to describe this classical approach here. Erich shows that this classical result can be found in Italy, but only in the south. Because the south is also known for forging test scores, Erich and his coauthors suspect a link between the two phenomena. They actually show that a decrease in class size seems to improves test scores only by an increase in test misreporting. I do not remember the proposed mechanism, but the result is counterintuitive at first sight (smaller class is easier to grade and thus misreporting is less of an issue, though active cheating is easier). David Jaeger showed evidence from an experiment that halving class time only decreased test scores in a university introductory economics course by a few points, suggesting that class time could be reduced without strong detrimental effects on learning. Stefano Verzillo showed evidence that Italian academics seem to react to the competitive pressure exerted by peers at the time of promotion: the more competition they face, the more they work if they are already good. The reverse is true for scientists with a lower publication record. I apologize to Rainer Winkelmann and Winfried Koeninger for not relating the content of their talks.