THE DIGITAL PROVIDE

THE DIGITAL PROVIDE: INFORMATION (TECHNOLOGY) MARKET PERFORMANCE, AND WELFARE IN THE SOUTHERN INDIAN FISHERIES SECTOR

R. Jensen

Quarterly Journal of Economics (2007)

Principal Research Question and Key Result	Do improvements in information availability impact market performance and welfare? Specifically in the context of Indian fisheries, did the introduction of mobile technology reduce the variance of fish prices in different regional markets, and did this impact welfare? After phones had been introduced the law of one price declined in violations from 50-60% to almost zero, whilst at the same time fisherman’s incomes increased by an average 8% and consumer surplus for sardines increased 6%.
Theory	Where goods are more highly valued on the margin in one market than another, a price differential occurs and profit seeking suppliers reallocate goods to that market, which eventually reduces the differential by reducing the price, increases welfare, and ensures that in the long run prices are the same. This is such that the law of one price holds, meaning that the only long run difference in prices for a product are consequent of different transportation costs to different markets. When fishermen are at sea, they have no contact with the markets they may eventually sell in, and they only observe the size of their catch, not of others. The fisherman can go to his local market or can switch (at a transportation fuel cost) to a nonlocal market although due to the cost they can generally only visit one market per day. They would go to a nonlocal market if they believed that prices there would be such as to get them a greater profit net of additional transport costs. However, all boats face the same dilemma, and as the price in any market is a function of the amount of fish caught, and each vessel only observes his own catch, then there is no way of knowing where prices will be highest. This means that in general they will sell only in their home market. This means that given that there is no ability to store the product (due to cost), and limited arbitrage by land (due to poor quality roads), the amount supplied to any market is almost exclusively determined by the amount caught in that locality. As catches vary widely in quantity, this translate into supply and hence a large variance of prices. Introducing search technology (phones) allows arbitrage to occur as the fishermen can call the markets whilst still at sea in order to discover where they should best land their catch. This should reduce price dispersion, and this is essentially what is being tested.
Motivation	Some people argue that ICT investment should not be a priority for developing countries as they face bigger problems in terms of education and health etc. However, this overlooks the fact that ICT can affect product markets, and many in the developing world are dependent upon sale of their output for their survival, particularly those engaged in agriculture, forestry and fishing. This paper looks at the link between ICT and one such product market, that for Sardines in Kerala.
Data	Data come from Kerala’s three northern coastal districts where mobile phone technology placement was staggered (allowing for exploitation of time variation) and masts were placed in the coastal cities, allowing fishermen at sea to use the technology. The surveyed 20 (random) fishing units for each of 15 beach markets every week for 5 years, every week, and obtained data on quantity, price, where it was sold, time of sale etc. The sample is then broken into 4 periods, one where no one had phones, 1 where region 1 had phones…. Etc. There are control variables based on wind and sea conditions, price of fuel etc.
Strategy	This is a difference in difference strategy with change in region on the left, and dummies for region and period, and region*period on the left with controls.
Results	Even the summary statistics are pretty remarkable; All fisherman fish in local area close to 100% of the time, in all periods. All fisherman sell in local market when no one has phones, 65% of period 1 fisherman sell at home after they get phone with all other fisherman continuing to sell at home…. Between – 58-70% sell at home when they all have phones. The pattern is striking. The same goes for the summary stats on price dispersion. Price dispersion is wide and falls dramatically by region as they acquire mobile technology. The amount of wasted fish also falls to almost nil in the regions as the get the technology. Before they get to the DID they pool the data and get a reduction of 38% of the standard deviation of prices. They then move on to their separate treatment effects regression (DID). They estimate the effect of mobile phones in region 1, using region II and the control, and then using region III as the control, and do this for each region, giving 6 resulting coefficients. Whilst there is some magnitude difference, the resulting coefficients range from 0.35-0.46 which is a pretty tight spread very similar to the pooled results, and they indicate a -4.7 to -5.8 reductions in local currency of the min-max spread i.e. a very large reduction in the variance of prices. Waste is also greatly reduced. The average price of fish declines, which gives the consumer some surplus, but due to less waste etc. the incomes of the average fisherman increase 8%. These changes are permanent over time, not one time effects.
Robustness	In order for the estimation to be consistent, the regions should have exhibited similar trends in dispersion of prices but for the introduction of mobile technology. He provides graphical evidence that the only shock to dispersion of prices occurs when phones are introduced which adds credibility to this parallel trends assumption. He makes an extremely detailed case arguing for the identifying assumption. He shows that the only dispersion shock occurred when phones were introduced. Also, if phones were placed in areas where there was good growth occurring (affecting fish prices) then there should be some effect noticeable before the phone service was actually switched on, as the time in between deciding to put in a phone network and the time it is switched on is substantial. He also shows that there was no significant change in the composition of the fishermen which rules out the possibility that migration was driving the results. He rules out alternative explanations: Phones increased profitability of fishing, and hence the number of fishermen – there is no correlation between number of units and the introduction of phones in a region. Phones affected variability of catch as they could inform each other where the fish were biting – regression of amount of fish caught per vessel shows no relation to size of catch and introduction of phones. Increased wealth of region due to phone could affect prices – he estimates using household data, the elasticity of sardines with respect to income and it is only 0.12, so unless the wealth effects of mobiles was huge (not to mention nearly instantaneous), then this could not explain. Phones could have altered the time that fish were sold, as fisherman chose when to arrive to exploit potential in their local market – he varies the time condition to include prices at all times of day (it is only for one hour in the morning in the main specification) and no significant change is observed in the coefficients. Phones enabled greater price collusion – this is not tested empirically, though local NGOs and the fishermen themselves strongly deny that this occurs. Phones in the hands of customers may have reduced the likelihood that they go home with no fish due to no stocks, and as such this acts as an insurance on leaving a market empty handed, thus reducing the variability of prices as customers are less likely to pay above the odds just to be sure to obtain some fish. In interviews, this does not seem to be happening. He gives GPS to several boats and thus creates a cost variable to the different markets based on their fuel usage. He uses this to test when price spreads have violated the law of one price for the sample. He finds violations in 55-60% of the observations in the pre phone era, and close to zero for the post phone era.
Problems	This paper is really thorough. External validity is of course a problem. One of the very specific features of this setting is that the goods are perishable meaning they could not be stored, and this also made arbitrage by land very hard. For products where these are not issues less instantaneous forms of information gathering may be sufficient. This also indicates that how transport infrastructure interacts with the markets in question will determine how productive investments in these information systems will be.
Implications	Information is critical to making markets work, and improving market function can improve welfare. Interestingly these projects were private, so they do not rely on donor funding nor do they crowd out other investment. The project is self-sustaining as it is undertaken for profit. Thus there is an interesting role for private sector investment in improving market and hence development outcomes.

 

 

THE EFFECTS OF RURAL ELECTRIFICATION ON EMPLOYMENT

THE EFFECTS OF RURAL ELECTRIFICATION ON EMPLOYMENT: NEW EVIDENCE FROM SOUTH AFRICA

T. Dinkleman

American Economic Review (2011)

Principal Research Question and Key Result	What effect does electrification have on the ability of individuals to use their resources in labour market production, and what mechanisms are at work? The results indicate that for women in South Africa participation in the labour market increased at the intensive and extensive margin. Specifically the number of women working increased by 9 – 9.5%, and the number of hours worked increased by 3.5%.
Theory	Electrification could have various effects on labour supply. As electricity improves productivity in home activities such as cooking then it is possible that electrification could induce individuals to substitute away from the labour market toward those activities, thus decreasing labour supply. This is effect is likely to be limited given that the amount of meals etc. that can be prepared is generally bounded at fairly low levels. Electricity also increases the effective length of the day (through electric lighting) so may increase the demand for consumer goods and this may induce an increase in labour supply in order to fund those increased purchases. Electricity may also increase work opportunities in local areas by stimulating growth of new firms, and it may also create jobs within the household by enabling the production of new goods and services for the market (microenterprises etc.). This effect could dominate even in the absence of an increase in the number of formal firms. This paper looks at the effect that electricity had on home production activities, market employment, market wages, and migration.
Motivation	About 1.6 billion people lack access to electricity. Infrastructure in general is often invested in by national governments and development agencies, but we do not have a clear idea of what the effects of such investment are.
Data	The data are from South Africa which greatly increased its electricity infrastructure in the 1990s with a priority of reaching disadvantaged communities first. This objective was tempered by the costs associated with creating these new power lines (and also political factors), a fact that will be exploited in the IV strategy. The data are in panel format, from 1996-2001 with spatial data on the placement of power infrastructure in a sample of rural and ex-homeland communities. A second strategy uses individual level data on work, wages etc. drawn from four cross sectional surveys from 1995, 97, 99, and 2001.
Strategy	Given that the placement of power lines was non-random, a simple regression would be biased by correlations between variables associated both with power line placement and the outcomes of interest. Thus an IV strategy is used. Gradient of the land is an important consideration when thinking about the costs of creating power infrastructure, and so the gradient of the community is used as an instrument for the electricity network expansion. The importance of doing this is obvious when it is shown that non-electrified communities are poorer, have fewer adult women, more education, and are nearer to towns and roads. Instrumenting then for the arrival of the power network she regresses changes in employment (by men/women) on the electricity dummy, a vector of community covariates (density, education, sex ratio etc.), with community and district fixed effects.
Results	The results are not hugely compelling. The OLS estimates for changes in women’s employment are all close to zero and insignificant. In the IV regression, the full specification indicates a 9.5% increase in female employment although this is only significant at the 10% level. For men the OLS regressions are negative and weakly significant, but significance and sign change for the IV regressions. Using the household surveys to examine employment, hours of work, hourly wage, and monthly wage using an OLS and then a fixed effects model, reveals vaguely similar results, but significance is weak to non-existent in all cases, and the signs are all over the place. Evidence is presented on the mechanism art work. Average electric lighting increased by 23% and reliance on wood fuel decreased 3.9% and cooking with electricity increased 5.6%. The coefficients are much larger when electricity is instrumented for with gradient. It is suggested that the results on female employment are therefore due to a freeing up of time in the home, and as this is most pertinent for women, this is why we only find significant effects on the female labour outcomes.
Robustness	In a regression of the potentially biasing covariates (poverty, distance to road, educ) on gradient alone, she finds no significant coefficients which adds some limited plausibility to the exogeneity of the instrument story. A placebo test is conducted using areas that already had electricity in the period, and no significant results are found which is indication that results presented are not just due to spurious correlations. A test is done to see if migration is driving the results. I was so bored by this point I had to leave it.
Problems	In the OLS and IV regressions evidence is pretty limited. Female employment changes are small and insignificant, then large (positive) but weakly significant, and men are large (negative) and significant then small (positive) and insignificant. There does not seem to be a strong theoretical justification for this. The author claims that census data may be measured poorly for men, but it is not at all clear this should not also be the case for women. In any event, the estimates indicate that 15,000 new jobs were created for women in the area. Whilst this may be a fairly large number, as a result of all of the post-apartheid programmes, it is estimated that 2,000,000 jobs were created, and hence these particular jobs only represented 0.75% of those jobs (3.75% when weighted by population), which is not a hugely compelling result. For the IV strategy to be successful, conditional on baseline community characteristics, proximity to the grid and economic centres, and district fixed effects, land gradient cannot affect employment though any other channel other than the electricity network. However agricultural outcomes could be affected by gradient and they could affect employment. Whilst the author acknowledges that only 10% of individuals in the sample are involved in agriculture, this does not entirely solve the issue. This is because the instrument is shown to be very weak in the first stage with an F-stat on < 9 in the fully controlled specification. One of the key problems with weak instruments is that even a small departure from exogeneity of the instrument will result in large inconsistencies in the IV estimate, and it is not easy to discern in which direction the bias operates. Even if the IV strategy is good, as is noted in the paper it is measuring the Local Average Treatment Effect which in this case the effect of electricity for those living in relatively flatter areas. If people in those communities can better afford electricity once it arrives, or face lower commuting costs (and therefore higher net wages), then the effect is overestimated with regard to the Average Traetment Effect. This is unfortunate, as we have no idea of how much bigger that the ATE the LATE is. Ideally you want the LATE to be a lower bound estimate for the ATE. As far as I am concerned the only evidence that has been presented is some weak evidence that women’s labour outcomes improve, and separately that people cooked with electricity and had lights. I do not see a necessary link between the two, and this is not well established in the paper. Thus the mechanism between electricity and labour outcomes is not well evidenced.
Implications	Difficult to tease out implications for this one. Infrastructure may be important for labour outcomes, but it is difficult to get to the effects empirically. The results presented here are not compelling, especially as they account for only 0.75% of new jobs created in the period, and this may well be an overestimation. What is clear is that different people will respond in different ways, and that the structure of both the existing social construct and labour market will be important in determining outcomes from such projects. It should also be noted that in seeking to generalize these results to other contexts, that post-apartheid South Africa is a very specific setting and the results (even if solid) may not be appropriate for judging what would occur in other settings, and additionally, that is such a programme is rolled out to entire nations there may be important general equilibrium effects that this paper does not capture that may lead to different outcomes.