 Thank you for the organizers for the opportunity to present this work here. I'm going to present two pieces of work today. The first is brand new. It will come out as an IMF working paper within the week or two. And it is empirical work on the future of the world oil market. And the second part will be about a slight extension of the wheel work that I presented here in October last year that's using the IMF's global economic model to think about different scenarios for the future of the world economy. Here's the usual disclaimer. So by way of introduction, when it comes to predicting future oil prices, conventional models have great difficulties with that. And when I say conventional models, I mean econometric models that are out there and have tried to forecast oil prices, also surveys of expectations, futures markets, they cannot consistently beat a random walk. And a random walk says that the best predictor of tomorrow's oil price is today's oil price. So that's not a very good forecasting performance. And part of the reason for that is that econometric models tend to pay very little attention to oil supply. That has two problems. First, they cannot capture temporary oil shocks very well. But more importantly, especially at the current juncture, they cannot predict an upward trend in the oil price because these models tend to be demand-based and mean-reverting models that will always predict that the oil price goes back to some level and to have a trend that persists over a decade, which is almost what we have by now, is hard to explain for such a model. The alternative is to take the geological limits to the oil supply more seriously than these models do. And then the oil supply constraints are front and center for determining the future of oil production. But then in that case, there's no role for oil prices. And as I will show you, this is also not consistent with recent data. So what is needed is something that's in between. And the model that I will present in this paper combines both approaches, where geological limits will turn out to be very important to explain the upward trend in oil prices that we're in the middle of. But demand shocks through higher prices, by causing higher prices, can increase production as well. And that is also something that we have observed, as I will show you. Because this is a statistical model that we evaluate with state-of-the-art techniques, we can then say something about the uncertainty about shocks and also about key parameters that we can then directly evaluate. A few words about historical forecasts of world oil production. The point of the exercise is I will show you that the mainstream view has tended to overestimate where oil production is going. The peak oil view has tended to slightly underestimate where it's going. And these have been coming together like this. They're still very far apart, but they have been coming together. And so here are the EIA's projections made between 2001 and 2010. And it's a bit hard to see because there's so many of them, but it's an almost continuous downward revision year by year by year. These are projections out to 2020 with different starting points. These are the different years in which the projections were made, an almost continuous revision by a very large amount of more than 25 million barrels a day in 2020. So these are very large numbers. This started in 2003. What happened in 2003? In 2003, the official OPEC spare capacity took a nosedive. This was in early 2000. This was around the time of the Iraq War. And that triggered the beginning of this upward spike in oil prices that lasted all the way to the beginning of the Great Recession. It then dropped. And this was because there was demand destruction. And spare capacity went up. But now look at what's happening. Spare capacity is trending down again. And oil prices are trending up again. And so that's the picture as far as the official view is concerned. Here are three of Colin Campbell's forecasts. It's a slightly different aggregate of oil. It's regular conventional oil. These are the actuals again. And these three lines are three different forecasts made between 2003 and 2010. And again, until 2020. And there's a slight upward drift in those forecasts as well. Not as well in those forecasts. So this goes the other way from the official view. Now Colin Campbell's work is very much bottom up looking at individual oil fields with a lot of engineering knowledge. There's another approach that proponents of the peak oil view have used. That's a curve fitting, the DeFe model. That was done in 2005 with data up till 2003, predicting this curve here and this production profile here for future oil production. What happened? Because we just saw the data in 2003. What happened from that year onwards is that oil prices jumped up. And look what happened. There's significant deviations from the phase forecasts from that line that have lasted to the present day. But these deviations, nevertheless, have not led us to resume the historic growth rates. They have kept us on a plateau. So we want something in the middle of this. And so we built a model that I do not have time to go into great detail about. The only thing that I want to say is that we basically built a model with demand and supply curves for the world oil market where the demand curve specification is very conventional. And then there's another set of equations for potential GDP and the output gap, excess demand. All that is conventional. The supply curve is not in that it combines the DeFe specification of the peak oil view, which is called habit linearization, with price sensitivity of oil supply, where higher prices can make the oil supply go up over time. And that is basically the combination of both views. And have a look at what that does to the forecast accuracy of our model. And so here the table, I'm sorry, it's a little bit small. I should have made this bigger. But I'll tell you what's happening in words. These are out of sample forecasts for our model versus the closest competitor out there. And these are one year, two year, up to five year ahead forecasts. And what is shown here is the root mean square error. So when those numbers are large, it means the forecast is bad, relatively speaking. And what we're seeing is that for the real price of oil, our model does better than the random walk at any horizon. And especially at longer horizons, the forecast error is a quarter of the random walk. And as I told you earlier, the random walk is something that conventional models have difficulty beating. For oil production, these are the EIA forecast, root mean square errors. And these are our models. And our models are about half of what the EIA's forecast errors are. And so, again, on oil production, the model does very much better. I'm not going to go into the GDP numbers. Here's a little story. This is a decomposition of history where we're saying, let's start in 2002. Let's run our model forward from that point onwards. And this red line is running it forward without any shocks. And this blue line is running it forward with the estimated shocks that we estimated. We're using Bayesian non-linear estimation techniques, high-tech stuff. And this basically is the model with all shocks that fits the actual evolution of oil prices. And these blue lines and these other subplots decompose the contribution of different shocks to what happened to overall oil price deviations from this trend. Now, the critical thing is that this trend, even without any shocks, is up. And that is because of the Hubbard feature in the supply curve, meaning that even if no oil demand or oil supply shocks hit this economy, on the basis of estimates up to 2002, this model would have predicted an upward drift in oil price because of this peaking nature of the supply curve. But then, there were significant deviations. And where did they come from? Well, there were oil demand shocks in the early part of the 2000s as the economy was doing well and also output gap shocks a little bit towards the later part of that period, but mostly oil demand shocks as the world economy was booming. But then from 2005 to 2008 and also more recently, the last one or two years, oil supply shocks explain a lot of the action in the actual oil price relative to the trend. And that, what happened in 2005, we reached the plateau in world oil production. And we've been on it ever since. And so this is beginning to have a big additional effect. And output gap shocks, the great recession led to demand destruction. And that is largely explained by output gap shocks. Let me skip over this one in the interest of time. So that was forecast, out of sample forecast in history, now what we're doing is, let's take history as given and let's say, what does the future hold? So we're saying, we're in 2011, we have estimated this model up to here with uncertainty about shocks and parameters, et cetera. What does the future hold? Well, the point forecast of our model over the next, this is actually nine years, not 10 years, but roughly 10 years, has output growth of roughly 0.9% per annum. And if you look at the very latest EIA forecast, it's in the ballpark. That's more or less what they project. But I want you to also notice that there are very wide error bands around this. They reflect our uncertainty about what are going to be ultimately recoverable reserves and elasticity of demand and supply. And the flat output path for the next 10 years is within the 90% confidence interval. But let's not concentrate on that for now. Let's focus on the next one, which tells us, okay, what oil prices are going to be necessary to make this happen, to make this 0.9% output increase happen. What the model tells us, again, with great uncertainty because there's so much we do not know about the oil market, but our best guess is that the real oil price has to nearly double over the next decade. It's a little less than double on the point forecast, but it is in the ballpark of having to nearly double. And this is because we're always working against this geological constraint now that wants to drive production down. And if demand is to be satisfied through supply, it can only happen through higher prices that have to be quite significant. And depending on if you have, for example, very much lower elasticities, you would move to this part of the confidence bands to even higher oil prices. And given the historic performance of our model in forecasting oil prices out of sample, this is something that we are taking somewhat seriously. Conclusions, the objective was to evaluate a model that encompasses two diametrically opposite views of the oil market, the resource constraint view, and the view that prices are ultimately decisive, which is the economist view. The model performance, in terms of history, our decomposition into trends and shocks is very plausible. It tells a story that intuitively sounds right. For forecasting, we have a far better performance than competing models. What do the forecasts say? First of all, the EIA's latest forecast of 0.9% annual supply growth may be feasible, but the real oil prices would have to nearly double over the next decade, but that's subject to very large parameter uncertainty. And then there's one very important thing here. The effects on GDP are here in this model, and I didn't discuss them very much. There are negative effects of higher oil prices on GDP, but they're not enormous. They don't make the economy collapse. And that's based on historical data, where it's kind of hard to argue that that has ever been the case. There have been temporary blips, but to say, we have very little to go on in the data that would tell us a very persistent supply constraint would lead to a big collapse in GDP. We don't know that, right? And our model tells us the effects on GDP on the basis of the data are not too dramatic, but we think that there may be non-linear effects on GDP once oil prices breach a certain pain barrier, but we don't know what that is. And so our future research will be very much on that question. Now, that's where I'm connecting to the WIO that came out last year, and I'm going to present a succinct version of the simulations that were in there, because what we're doing there is we're basically using a model laboratory style environment of our global economic model that we're always using for policy analysis, not for forecasting, but for policy analysis, scenario analysis, et cetera, of which I am responsible for the theoretical development and the latest theoretical development is to put oil in there. And so we need to understand two questions about oil. First, what is the outlook for production and prices? And I just talked about that, and there's more work to be done here, but we have at least some idea now. The other question concerns GDP, and I talked about that at the end of the previous paper, how might oil scarcity effect output? The WIO gave some very tentative answers, but we feel that we'll need a lot more work. So a brief overview of the GMO, the IMF oil version. GMF is a choice theoretic dynamic business cycle model. As I said, it's very frequently used at the IMF for all sorts of purposes. I only want to talk about how the specification that I'm going to use today differs from a version without oil. We have an oil demand function where there is a very low short run and long run elasticity of substitution between oil and other factors of production, a low cost share, because that's what the data say. This is also what the data say. The output contribution of oil is equal to the cost share in the baseline, but we then experiment with a version where oil makes actually a very much bigger contribution to output than its cost share would indicate. And there is a justifiable, academically justifiable specification of aggregate production in the model that makes sense of that assumption. For oil supply, we treat oil as an exhaustible resource whereby it's mostly an exogenous endowment, but there is a price elasticity of supply of 0.03, so that's low. The price response is low. We have five scenarios. First one is reduction in oil output growth by 1% per annum relative to trend. This would be going from 1.8% per annum to 0.8%. The alternative one is to have these larger output contributions of oil. So if I don't have enough oil, it makes much more difference to how much output I can produce. Alternative two is to have a reduction in oil output growth by 3.8% per annum, which would take us to an absolute decline of 2% per annum, which is sort of like the more pessimistic peak oil scenarios that I've seen are more or less that order of magnitude. Then, and this is the wheel up to here, and just for the sake of argument and sort of to drive home what's at stake, for this presentation, just combine alternative one and two and to go through what the model tells us what would happen there. And then I've also combined this alternative three with a zero elasticity of substitution between oil and other factors of production, just to have an extreme case to illustrate what would happen. Now, so here we have all these scenarios in one graph and from the baseline to it gets worse as I go down this list. And what we have here is GDP absorption and current account for oil exporters for the rest of the world. Here we have the oil supply, the price of oil and the real interest rate. And so first of all, the output effects range from 0.2% per annum less growth to over 2% per annum in the worst case scenario. But these scenarios are really much worse than what's in the wheel. The worst one that was in the wheel had a 0.75%, I believe roughly, annual loss of growth over these 20 years in the projection. So that's one thing. Current account effects in the oil importing countries for less than 2% of GDP in the baseline to 10% of GDP in the somewhat crazy last scenario that I'm putting out here. So I don't want you to take that too seriously. This is just sort of playing with the model and say, if there really wasn't a substitution possibility between oil and other factors of production, how bad could it get? And next, the price of oil. The baseline for the price of oil, which is that blue line, is actually relatively close to what I've just shown you in that empirical model. 100% after 10 years and another 100% after 20 years. For different reasons, the models are not strictly comparable, but it sort of gives me comfort that I might be in the ballpark with the baseline. But those are already very large numbers for the increase in oil price. This green dotted line is what we also had in the wheel, 800% after 20 years. That is a number that is so large that I think, in my mind, there is no doubt that there must be some nonlinearities for output, but the model doesn't have those nonlinearities. So the model just basically says, well, there's an output effect, yes, but most of the effect of oil scarcity is sky-high prices. If you had a link from these sky-high prices to more of an effect on output, then you would get demand destruction, destruction of demand for oil on a larger scale. The final point, real interest rate, don't even look at the bottom two lines, they're just too crazy. What would happen is when this large increase in revenues to oil exporters happens, we assume in the model that they accumulate that in an oil fund, that they don't spend right away. This would lead to a very large increase in world savings, and that increase in world savings for given world investment would drive down the world real interest rate to a point where the real interest rate would become so low that nominal interest rates would have to sink too close to zero, meaning monetary policy would be constrained by the zero lower bound on normal interest rate. So apart from leading to all sorts of current account imbalances, it would also lead to problems for monetary policy if the worst case scenarios were to happen. So the conclusion, the oil price effect in the benchmark is similar to our empirical paper. Output and current account effects in the benchmark are modest. The alternatives introduce assumptions that could lead to much larger effects, but these are at this point, they are just sort of, we're just playing in a laboratory style model environment and we need to link that to data more and that's what we're starting to work on. The output effects get larger but are still not enormous but the oil price effects are enormous under those alternative assumption and the key question just like at the end of the previous paper is at what level does the output response to higher oil prices become non-linear in the real world because in the model it is approximately linear and that's all I have. Thank you. Thank you. Thank you.