 Of I'm going to be talking about today about agricultural commodity price shocks and the effect on the growth in Sub-Saharan Africa. Now, my work on commodity prices has mainly been on the univariate study of commodity prices. So looking at trends cycles, volatility, and persistence. When I was here on sabbatical recently, Tony and I discussed about doing some work together on looking at how commodity prices can affect growth in sub-Saharan Africa. Yr ystafell i gael g critically y prysiau yn gym stackedwyd anhyfodol y Llywodraeth i Llywodraeth Aelodau. Nid yw weithio dod gen i'w gwybod, mae rôl yma, unrhyw arall, fod rhywunio'r gweld o'r datblygu, i wneud arferiaeth sydd o gwybod iddyn nhw'n ei wneud flasig o'r ysguodadau neu rhywun i'w gwybod i'w cyffaint o'r Llywodraeth i Llywodraeth. So what we're going to focus on here is commodity prices because this has been singled out as one of the primary drivers for income volatility in Uppsaharan Africa. The problem with commodity prices is that they are becoming increasingly variable, especially since the 1990s, they are becoming increasingly variable. They are broken trends in commodity prices. There is persistence, which is dynamic in nature. Oes ddifwneud ti yn ymweldio eich ff waysawe i yw'r dweud, sighs rai'n gynnig i'r bywydol ag oedau ymwyng rich, i'w sgwylio'r sgwylio yn cymhloedd cael eu rai, o fe wneud i yw'r syniad yn cymorth i ddechrau. Mae'r cymorth a newyddol yn fanch, ac mae'r cymorth i ddechrau yn fwy am yr hwrnd dechreu, ac mae'r cymorth i ddechrau yn cymorth i ddechrau. Mae'r cymorth i ddechrau yn cymorth i ddechreu, acolodol yn dwech yn cymorth i ddechrau It is difficult to deal with these sorts of things and of course this has impacts on aid flows as well, it just complicates the whole picture. So, there is a large research that has taken place on commodity prices as I said, trends, cycles, volatility and persistence and they are all found to change over time which makes it all the more difficult. So, when we start thinking about how commodity price shocks can affect economic growth in sub-Saharan Africa, we need to think about what particular manifestations of commodity prices might affect the incomes in these countries. So, what we do in this paper is we choose two particular manifestations of commodity prices, which have been used in the literature, but for different primary commodities. So, what we're going to focus on here is agricultural commodity prices, and I've basically used this article by Anderson and Bruckner, where they have done some detailed study by looking at Penn World Table data to find out that for the period of time that we're going to consider as well in our paper, that the average share of GDP in sub-Saharan African countries from agriculture has been a considerable amount. So, the motivation over here is to study agricultural commodity price shocks on per capita incomes, and particularly what we are going to be interested in is to separate the shocks in commodity prices, to divide them into positive shocks and negative shocks, and also within those positive shocks we look at sustained positive shocks and negative shocks as well. And then we'll do some simulations to find out if we shock this system how output changes over time, we'll look at a certain horizon, and we'll find out whether there's any difference between these positive and negative shocks. So, the question is what sort of price shock are we going to choose? Now, in the literature there are two types of price shocks that have been used for oil price shocks. One has been provided by Mark in 1989, published in the Journal of the Political Economy, where he divides oil price shocks, oil price movements rather, into positive and negative movements, and traces out the effect of those oil price movements on the US economy. And what Mark finds is that positive oil price movements tend to have a more impact on the US economy than negative oil price movements. And since then, this particular model by Mark has become quite popular in the literature on commodity price movements, especially oil price movements, and they have been used in the literature until only recently. Also, Kilian and Vigferson have been using this type of study quite a lot. Hamilton in 1996 and then again in 2003 refined the definition of this oil price shock by talking about a sustained increase in oil prices to be what he calls a net oil price increase or decrease, and what sort of effect that has on the US macroeconomy or other economies as well. And what they do is Hamilton chooses the current oil price and then measures it against a certain period of time. So if that current price is greater than that oil price over a certain short interval, then that would be considered as an oil price shock. So basically it's about commodity price increases that establish new highs relative to recent experience. So let's talk about some perceptions of commodity prices because these sorts of studies have been used for oil prices. Now if we talk about commodity prices in general, which include agricultural prices, prices of metals, and other primary commodities, some of the classic studies, and there are other studies as well, but I've just chosen some of the classic studies. Deaton and Lloroc in 1992 in their paper in the review of economic studies, and as well as Deaton in 1999 in the Journal of Economic Perspectives, have talked about the nature of commodity prices. They've looked at the basic statistics, and what they conclude is that commodity prices are characterised by flat bottoms punctuated with sharp peaks, and these sharp peaks that occur in commodity prices are driven by stockouts. So that's what causes this sort of movement in commodity prices. In a separate paper, which is more of a policy paper, Deaton and Lloroc, they point out that this uncertainty about commodity prices makes it all the more difficult and complicated for policy makers to handle these sorts of shocks. So what we plan to do is we plan to borrow these two types of shocks that are used in oil prices and use them for agricultural commodity prices, primarily because in the studies by Deaton and in also other studies, for example one study that comes to my mind is a recent study by Harvey, Kellard, Madsen and Warhur, they have used primary commodities, this whole set of 25 commodities over four centuries, and within their basket of commodities where they've looked at agricultural prices, which include food and non-food, they've also included oil. And I've also looked at some separate studies, I've done one study myself where I've looked at broken trends in commodity prices and again broken trends in oil prices, and they're characterised by the same sort of movements. There is dynamic persistence in commodity prices, there's also dynamic persistence in oil prices. So in terms of their statistical nature, they're quite similar. So what we have decided to do in this paper is to take these two manifestations of commodity price movements and which are applicable for oil prices and have been studied in the literature quite extensively, especially by Kilian and Vigfoson, and see what sort of effect they have on sub-Saharan Africa. So to give a literature review, again the literature is quite substantial and so if anyone of you has done some work on this, please don't feel offended if your work is not here. But this is just a snapshot of some of the studies which I found quite useful. Easterly in 1993 they published a paper in the Journal of Monetary Economics where they talk about growth regressions. They just model growth regressions and they find that commodity terms of trade have a significant effect on output volatility. Mendoza, Cousin, Reisman, they use calibrated journal equilibrium models and they also find a similar sort of result. Blinion Greenaway, they use panel data and they find that for variations in output there is a negative relationship with output. Sorry, variations in commodity terms of trade, there's a variation in output and the relationship is negative. Blackman also uses a certain length of time but that's historical data from 1870 to 1939 and again they find similar results. Agion uses general methods of moments and again the results are quite similar that there is a strong relationship between commodity terms of trade and output volatility. Now Deaton and Miller in 1996 and Hofmeister and others in 1998 they use a VAR model and they actually find that the relationship is not so strong as previous studies tend to suggest. But then Broder, Radatz, Collier and Gaudaris in 2012 have used panel VARs. Collier and Gaudaris actually use a panel ECM to find out the relationship is actually quite strong between commodity prices and economic growth. But there are some drawbacks in these studies and what I'm tempted to pick out is that the dynamics that they assume among these cross-sectional units are relatively homogeneous. That is not quite true because there are studies which show that the countries can be quite different, quite heterogeneous in nature. Now Radatz does acknowledge that and he just sort of simply states in his paper that low income countries tend to be relatively homogeneous. We're not mixing up low income countries with middle income countries but nonetheless it's a fairly strong assumption and that's acknowledged by Radatz. And if this assumption fails according to Pesseron and Smith on the impact of the shocks on economic variables it will be inaccurately estimated. The other explanatory variables that they use in all these past studies is a single commodity index which is the Deaton and Miller index which is a popular index which is used in studies. And what we have seen in the literature, especially the current literature on commodity prices is that these commodities behave quite differently. Even among categories such as beverages, coffee, cocoa and tea for example, they behave quite differently. They have different dynamics. Even metals like zinc, copper and lead, they have different dynamics. So to put them together as one commodity index that has been done in past studies seems to me to be a fairly restrictive assumption. And also that's been, I've just recently found out a paper by Eucelius Moller and Tarp in Oxford Bulletin of Economics and Statistics where they have lucidly listed the advantages of using time series methods over other methods such as panel and cross section. I'll quickly move through this. This would be the standard VAR model that would be used if one were to look at a linear approach which has been used in past studies. All past studies which are used in panel VAR and panel ECM have used a linear approach. So basically what we have over here, you've got the B21 coefficient in the second equation for example. If that is found to be significant, those coefficients, then you could say that X could have some causal relationship on Y. So X could be commodity prices. Y could be GDP per capita. Right? Now what we do over here is we introduce this extra term which is XT+, which is basically the model of MOC. And this is not what MOC actually did. He had a single equation model. What we are doing over here is we are borrowing the model of Kilian and Vickerson which was published in Quantitative Economics 2011. So we are using the same type of model that they have used. So that XT+, it's got the coefficient G21. If those coefficients are significant, then simply that means that the XT+, which actually senses out the positive changes in commodity prices, they can have a causal effect on GDP which is measured by Y if those coefficients are significant. And in that way B21 coefficients in the second equation would pick up the negative shock. So you're separating out the positive and negative shocks. Now Hamilton uses a slightly more complex model where you have XT+, which simply means a sustained increase in commodity prices which is measured by this max function over here and in which is set equal to three. That's the typical period that Hamilton uses in his studies. And we're going to use that same level as well. So again G21 measures that positive sustained effect on GDP. And then we compute the impulse response functions if we find that there is asymmetry in these types of VAR models and that simply is calculated by looking at one standard deviation shock which could be positive or negative, which is denoted by delta and H is the time horizon, which we choose in this data set to be five years. So the data that we used is GDP measured in constant prices. I'm going to quickly go through this. We use the Griliang index. This is a very popularly used index in commodity price studies. The sample period is 1960-2010, simply based on the sample size that's available. And we carefully choose nine agricultural commodities that are closely related to the GDP per capita that is available for various countries. So just to give some idea, a set of graphs over here and as you can see as Dieter and Miller suggest, you've got these sharp spikes which can tend to be persistent at times. Some basic statistics about the studies. There is a fair degree of persistence which is picked up by the autoregressive parameters AR1 and AR2. There's a fair amount of coefficient of variation. The skewness is positive, as we would expect. The number of positive spikes outweighs the number of negative spikes. And again, I'll just quickly whiz through this. You've got different countries listed over here and you've got the commodities which I found out from the FAO database and other related studies where there's a fair amount of dependence just on one single agricultural commodity. Now, these are the tests that we have over here. If we were to look at the Mark model that we talked about, you can see that there are four different countries which seem to be affected by this asymmetric shock. So if there are positive shocks, you can see that you can separate them out from the negative shocks and they tend to be quite significant. In the case of the Hamilton model, there's only evidence of two cases, but one of them, that's Cameroon, when compared with a linear model which is given by the last column, the linear model gives a better fit when I use a model selection criteria. So in the case of the Hamiltonian-type model, you've only got one particular case which is cote du va. But nonetheless, it's a very important country because as we saw from the previous table, a fair amount of dependency on one single commodity which is coco for the case of cote du va. And then given that we find that there are five countries which have found to be asymmetric, I've included Cameroon as well just as a comparison. We do this impulse response function, so it's an innovation accounting exercise, and we calculate the responses due to positive and negative shocks and what happens is we find out whether the differences are significant enough. So the graphs would look somewhat like this. I'll just quickly show them what the graphs look like and then we find out whether the differences between these graphs are significant using these p-values. And as you can see that in the case of Chad, cote du va, the Democratic Republic of Congo and Malawi, they're quite significantly different, but then over time some of them tend to be not so different but sometimes the difference persists. So finally, thank you. The last slide, you find that there is, if you take the linear case which has been chosen in most studies, there are about six countries which show that there's a causal relationship between commodity prices and economic growth, but we find that just as many countries can have an asymmetric response and that complicates the whole policy-making issue because then you've got these positive and negative shocks which can be quite different and given that countries are advised to accumulate reserves doing the good times and spend them doing the bad times, setting the school can be quite difficult given this evidence-based symmetry responses that exist. On this complicated note, I'll finish it. Thank you. OK, thank you very much.