 OK, so good afternoon, everybody. The paper I'm presenting this afternoon is joint work with Sukti, chair from the Cambodian Economics Association. And we're going to talk a little bit about clustering and looking in particular at the possibility that what we're observing in the data are spillover effects or competition effects. So just to give you an overview, because I don't get to the end. So what we're doing is we're investigating the pattern of clustering in the Cambodian context and microphones. It's better, maybe. So we're looking at whether or not we are seeing competition channels or technology spillover channels at work and whether they can explain the pattern of clustering that we observe. So the four main questions that we address are, first of all, whether firms are more or less productive, where there are clusters. Secondly, are the different types of firms impacted differently by the clustering, whether or not there are productivity spillovers and whether or not there are different types of firms are affected differently by these productivity spillovers. So the motivation has been outlined already by our previous speakers. So the geographic clustering of firms can impact on productivity in different ways by reducing transport costs, by allowing access to a common pool of labor, and then by allowing for technology and spillover effects. It also increases competition. And as a result, firms become more efficient through reducing slack or using their costs more efficiently. And so this competitive pressure can be another source of productivity growth as a result. There's an emerging body of literature that looks at this in developing country context. But the evidence is still quite scant. While in developed countries, it's been very well documented, particularly in the case of the US and the UK. There are some exceptions here. And the one, I guess, that's closest to this paper is the one by Ruslan Mons and Mulu and other authors, which looks at the case for clustering in Ethiopia. What I kind of want to look at first also as a motivation is why it should be given special consideration in a developing country context. And the two reasons I kind of say is that, well, first of all, it's already been given prominence in lots of industrial policy through the formation of industrial parks and through the formation of export processing zones. And there's not really been an evidence base for this. So that's one reason why we should take a closer look. And second of all, there may be different mechanisms at work compared to developed countries that are less well understood. So in terms of that latter point, why might the impact be different in developing countries? Well, first off, firms in developing countries potentially have a lot more to gain from clustering. They're starting from a much lower technological base. So the spillovers of new technologies, new innovations are likely to have a greater impact as a result. Secondly, competitive pressures might also be more pronounced in developing country contexts when we think about firms operating in a cluster, particularly at early stages of industrialization, and particularly where you may have underdeveloped physical infrastructure, which prevents firms from locating away from where they're selling their goods or services. And as a result, I mean, if they're located very close to their customers, there may be more of an incentive for firms to protect their technology to share less amongst each other. So in this setting, clusters might actually have fewer technology spillovers, but more competitive pressures as a result. This can prevent small firms from growing and maybe act as a deterrent for firms to locate close together. Also, the composition of clusters might be very different. And what we look at in this paper are service sector firms. So we look at service provision. They make up a very large proportion of small firms in developing countries. And their competitive pressures might be even more pronounced given that they must locate where they're providing the service in many cases. Also, informal firms make up a very large proportion of economic activity in developing countries. So in this paper, we also look at the impact on informal firms. So just to describe briefly the mechanisms that we're thinking about in terms of the competition effect, you would expect that the more firms that are located in close proximity, the tougher the competition will be, as we said. So firms should appear more productive in markets with more competitors as a result. And then the productivity effect, you might expect that firms experience spillovers from other firms located nearby. But this will depend on the characteristics of the cluster and will depend on the characteristics of the firm. So technology transfers can happen through the movement of labor between firms. So they bring with them what their knowledge that they've learned from the firm that they were operating in. This is probably more likely where you've got a lot of large firms clustered together, and it's probably more likely in high tech sectors. But you can also have spillovers through the actual copying of technology or the sharing of technology. So things like technological complementarities, where you've got firms that start to interact each other because they're located close to each other. One of them wants to introduce e-business, or one of them wants to introduce e-banking, and as a result, the rest of the firms also will do this. So there's complementarity in the technology. Sharing of technology is probably less likely for close competitors, given that they will have a greater incentive to protect their productivity advantage. Of course, there are lots of identification issues in trying to look at the impact of being in a productive cluster on individual firm productivity. There's three different problems in terms of identifying a cause effect. Well, first of all, firms may be more productive in large clusters because they've gone there because there are natural advantages. So they're close to good infrastructure, they're close to a reporter, or they're close to some kind of large market like a city. So that's one of the things we have to consider. Second of all, there's endogenous location choice to consider, more productive firms select into more productive sectors. So this makes the impact of the cluster itself difficult to identify, and then you have the reflection problem which makes separating out other kind of common shocks that impact everybody in the cluster from individual firm productivity. In this paper, we only have cross-sectional variation to exploit, so this makes the identification even more difficult. So what we do with our data that we do have is we try and control for each of these other factors, and then in the end, end up, hopefully, a convincing that we've isolated some kind of productivity spillover channel. The first step is to control for natural advantages. So what we do is we control for the density of firms within the cluster. So the larger the number of firms in the cluster, because they're more likely to be naturally advantageous areas, we reckon that we have controlled for that. Second, we try to isolate our competition effects. We do this by looking at the proportion of firms in the cluster that are in the same sector. This is evidence of being in more competition, which are close neighbors, and a positive coefficient would suggest that competition effects makes firms more efficient, if you like. So it's kind of an efficiency-enhancing competition effect. In cross-sectional data and the type of data we have, which is revenue-based, we may see a negative effect, and it's what we do actually see. And this is due to the fact that, you know, where you've got a lot of competition, you're going to earn lower profits. And reallocations, I guess, can happen at a lag. So there'd be sufficient firms that eventually, I guess, exit. And then there'd be a reallocation towards more productive sectors. So in aggregate, there may be improvements, but at the firm level, this kind of competition effect may be negative. To control for the endogenous location choice, we control for the average productivity of all other firms in the cluster. And this, I guess, captures whether productive firms locate in higher productivity clusters. And once we've controlled for all of these things, we isolate these productivity spillover effects using the average productivity of all of the firms in the cluster that are in the same sector. So having controlled for all of these other things, we hope that this variable here is capturing what we're interested in identifying. To control for common shocks, we only have a cross-section, but we do have some information from two years previous to that and enough information so that we can control for changes in the cluster between two periods. So we control for changing the size of the cluster and we control for the change in the proportional firms in the cluster that are in the same sector. We also compute each of the cluster level variables excluding the individual firm to further help with this reflection problem. So the model that we estimate is a cross-sectional model where we look at the output of the firm, which is revenue-based. We have our four measures there that capture each of the different variables that I've mentioned. The two controls for the correlated effects. We have our typical firm characteristics in there along with the inputs. We also include sector-specific fixed effects for each of the fixed effects and we look at it at the district and the commune level. We have a revenue-based output measure so it doesn't capture physical productivity, which is a problem when you're using these kinds of data. So it'll also capture prices and differences in markups. What we don't, we can't really disentangle that using this one cross-section. We do take it into account, the fact that the effects might be different to more competitive sectors than non-competitive sectors. We do take this into account when we're interpreting our results. The data that we use is from Cambodia. We have 2011 is the main data source, which is the Census of Industrial Production. It covers 500,000 establishments and we can match this to the enterprise listing, which was initially just a listing of firms with very basic information on employment that we use to look at the change between 2009 and 2011. It also contains the location of the firm so we know the village that they live in. We don't know the actual street number, the GPS code, but we do know what village they're in. Just to give a very broad overview of the types of firms we're talking about, most of them are very small. So it's kind of a different, we're looking at a different kind of clustering really here. We're looking at the very, very small micro-firms and how clustering is impacting on them. Majority or service sector firms, I think that's a nice contribution here because it's not really been looked at in too much detail before. Very few are formal, so only 8% of them are registered and lots of them are located in the home and only 1% are actually foreign owned. 15% of firms are located in urban areas, so there is a certain, there's a good deal of dispersion around Cambodia, but there is a lot of concentration of business activities within villages, about, you know, fifth of firms within a village on average are in the same four-digit sector. So it's just a couple of maps, we've only got five minutes, so this just shows the density, the darker, the more densely populated that area is with firms and in terms of the number employed also. So you do see a certain amount of clustering, particularly around the main urban centers in Phnom Penh and also just along the main railway line also there. So in terms of the results, so in the first step, we just look at natural advantages, this is controlled for using the number of firms in the cluster and it's positive, which is what we would expect. Then if we conclude our control for the proportion of firms in the same sector, and this is what's capturing our competition effect, we find that it's negative, okay? So this is a cross-section, so what this suggests is that the more firms that are in the same cluster as you, this has a negative effect on your revenue-generating ability and because we have a revenue-based dependent variable here, I don't think this is too surprising that it makes your actual revenues less. Now, the extent which this leads to reallocation effects, we can't tell in our data because we only have the one-time period. And this effect holds regardless of where do you think of clustering at the village level or the commune level and including different levels of regional controls. We disaggregate this by types of firms by including some interactions and we find that for the registered firms, the competition effect appears to be even greater. So this is a different group of firms in that sense. For manufacturing firms, it appears to be even greater. This is the interaction term here, so it's negative for all firms, but even more for manufacturing firms and the size doesn't seem to really matter in this case. And our core result, I guess, is looking at the extent of which we're observing productivity spilloers. So here we've controlled for all of the other things, the other effects, the correlated effects and all of the other effects. And we're trying to isolate the impact of the average productivity of firms in the same sector on the individual productivity. And we find very limited effects, really. So when we're closer at the village level and include our district fixed effects, there's no impact here. When we're closer at the commune level, which is the broader level of aggregation, we do seem to find some small productivity spillover. And what's also of note here is that this competition effect seems to have disappeared. This is, I guess, something we might expect in a village where the firms are located very close together. There may be a lot of competition, whereas when you think of a broader level of, a broader cluster, the competition effect is less and the productivity spillover effect becomes more notable. Disaggregating by the types of firms, we see that when we look at the registered enterprises and we think of a cluster as a commune, we find that this productivity spillover effect does exist, but the competition effect is far outweighing it here. When we look at the unregistered enterprises, however, the informal enterprises, it seems that this kind of clustering at the commune level is leading to productivity enhancing spillovers. The competition effect has disappeared. Looking at manufacturing and services, what's interesting here is that the services where at the village level, where they really are competing with each other, we don't find any spillover effects, whereas when we look at the broader level of aggregation of the commune, where they are less likely to be competing directly with the other firms in the network, and we don't find any competition effect, but we do find a productivity spillover. So it seems like the level of competition and the really determines the extent of which we can see productivity spillovers. For manufacturing firms, while there are competition effects, we do observe some spillovers here at the village level, but they're of a much lower magnitude. Small firms, we find a similar kind of story as you do find for the informal, or for the service sector firms. A positive productivity spillover with no competition effects. For the medium and large firms, there are productivity spillovers within these clusters, and this is a strong enough result given that there's no competition effect here either. We do some more robustness checks. We limit our analysis to firms that were in existence in 2009, so to exclude the possibility that they're selecting into this location because of the specific productivity of the time, and we find that our main result holds. So I think I'm out of time, so this is just really a summary of what we have found, and I guess from this, it's one to think about what the policy kind of conclusions might be. We see that there are observed benefits to performance from clustering, but they don't outweigh the negative effect of competition. It really matters here in terms of whether we see productivity spillovers, it matters the extent of competition between the different firms. So I guess introducing some flexibility, looking at why it's more difficult for firms to compete if they're formal, looking into diversification of customer base of firms, ensuring there's applied necessary inputs. All of these things might help to make these productivity spillovers realized. Sorry for wishing that at the end. Thank you, Carol, for a very interesting presentation.