A Fast Green Energy Transition is Likely to be Cheaper than Business as Usual

J. Doyne Farmer is Director of the Complexity Economics program at the Institute for New Economic Thinking at the Oxford Martin School, Baillie Gifford Professor in the Mathematical Institute at the University of Oxford and an External Professor at the Santa Fe Institute. His current research is in economics, including agent-based modeling, financial instability and technological progress.

In this presentation at Harvard Kennedy School, Prof. Farmer shares new research that shows that wind, solar, and other renewables would deliver energy security to the world and save $12 trillion by 2050.

Transcript

DISCLAIMER: This webinar transcript was loosely edited and there may be inaccuracies.

Farmer: It's great to be here. The work I'm going to talk about today grew out of a visit in two thousand and eight by the Director of the National Renewable Energy Lab, Dan Arvizu, who came to SantaFe Institute to ask us if we could help them think out of the box, and what were they missing? And during the course of that visit, and a workshop, it became clear to me that the costliness, and the correct path to dealing with climate change really depended on not what technologies cost now, but what technologies will cost twenty years from now. So that set me on a path of trying to understand, to try to produce reliable answers to that question. And so that's what this talk is about.

Now, just to provide a little context, energy makes about seventy five percent of emissions. So, if we can make the green energy transition, we've made most of the transition we need to make to deal with climate change, or at least to stop increasing Greenhouse gases in the atmosphere.

And the punchline that I'm going to deliver is that we can, and I’ll come back and discuss what "can" means, make green energy transition quickly in a profit, and I'll throw out a scenario at least a possible scenario for doing that discuss a little bit about the possible roadblocks to making that happen. Now, most of the story is in this plot that I'm showing here, which is giving the history of the evolution of the global energy landscape over the last one hundred and forty years.

And so, we see in the panel on the left, where we update on the x-axis, and cost measured in standard units on the Y-axis we see the price of various forms of making energy through time. For example, here's the prices of coal prices of oil, coal fired electricity. And so, you see they're kind of bouncing around, and one of the striking things about this is while they wiggle, there's no overall trend. In fact, I remember at this meeting in two thousand and nine. Dan saying, well, solar energy prices are coming down. And Dan saying, well, how do you know that coal prices aren't going to do that, too? And that actually ended up with a paper that James wrote on the cost of coal-fired electricity, showing that yes, it did actually come down for a while. But then it went back up, decomposing that and finding the causes. And you know the main cause is that ultimately cold, for electricity is going to money about coal, which, as you can see from here and comes down in mathematical tests, is essentially flat over time.

These be a very different week than these. So, you see solar energy, wind batteries, electrolyzers. And so those have been coming down quite systematically through time.

And at present we're in a peculiar state where we have many different sources of energy that are all competing around roughly the same cost. So, you can see it's very unusual. This over here shows history of energy, so you can see traditional biomass remaining roughly constant, coal becoming dominant, oil and gas rising through time, nuclear power rising and then flattening out, and then once again solar energy when batteries, electrolyzers all shooting up and race where they're being deployed at around forty percent per year.

So solar energy has been increasing deployment at forty percent per year. For the last thirty years it's been dropping a price in ten percent per year for the last thirty years, which is something unique in terms of historical energy supply. So, the question is, what are these going to do? Are they going to just flatten out and remain around the same price level with the other ones? Or are they going to continue to drop? And so that's what this talk is about, and then putting that together to think about the whole green energy transition. And, by the way, stop me at any point, if you have questions.

So, I’m just going to make a few observations about technological change. So, these are empirical observations, which is that first of all technologies, improve at very different rates, rates that are technology specific.

These rates are highly persistent, and this only becomes clear if you look at granular data, so you can really identify individual technologies and there's an identified identification question. We can come back to say more about if people are interested. But for now, let's say there's a such thing as a technology that you can track the cost of through time.

I'm going to argue we can use this to make predictions in one of several ways. One is to just note the trend. So here we take a span of over twenty years. The price changes in goods ranging from hospital services to television sets all starting at some standardized zero point, and we see that over those twenty years hospital services went up by two hundred and twenty five percent. The quality adjusted price of televisions went down by ninety-five percent. So, in other words, by a fact, television has dropped in price by the fact of twenty over those twenty years hospital services went up by a factor of two or two and a quarter, and so this illustrates the point about diversity. It also illustrates my point about persistence. Hospital services went up every year. Television sets dropped every year. So, these changes were quite persistent through time.

This illustrates the point about heterogeneity from a different point of view. So, we take the price indices for us, goods relevant for us, manufacturing from one thousand nine hundred and fifty, eight to two thousand and eleven. Group them by type, and what you can in this re-plot. Now make a histogram of the average annual growth rate set for the price, and you can see that most things like here is a valve middle that most things sit on, and most things are not too far away from just going along with the herd, the herds moving at the inflation.

But then there's some set of things. In this case, all computer electronics related things that are dropping in a cost in a much higher rate, suggesting that the heterogeneity of rates of change is a highly skewed distribution with a heavy tail on the positive side. We've done some other work that's not yet published that reinforces this point of view.

It's not just computers that do this, though they were one of the main things to do. So how do we take advantage of this? Well, we're going to make use of empirical laws for forecasting technological costs based on historical data. The most famous of these is Moore's law, so more in one thousand nine hundred and sixty-five made a pronouncement that semiconductor devices increased in density, on chips and identity, increased roughly, became twice as great every two years.

He later on, adjusted the timescale that I always mix up which way the adjustment went. But it's been a remarkably accurate prediction now over a seventy-year span. There's another, even older author, who is a very interesting figure. He is the brother of Seoul Wright, who is a famous evolutionary biologist, and Theodore Wright, who was considered one of the founders of political science, but he was the black sheep of the family who went off to World War. I became a flying ace and went to the aviation business, and in one thousand nine hundred and thirty-six he wrote his one academic paper, just mentioning that the cost of producing a specific airplane from a specific factory drops by twenty percent every time the cumulative production of that airplane doubles just an empirical observation.

And so this is an example of that actually, that I think James prepared for me a long time ago, and since then there have been a thousand papers written about Wright's law looking at all kinds of other technologies, and demonstrating that they did this and the law is also moved from being about a specific item from a specific factory to being about much broader items like solar photovoltaic cells rather than globally, rather than solar and photovoltaic cells produced by a specific factory. And that's the sense in which I'm going to be using here.

Wright, by the way, went on to become the head of an aircraft manufactured during World War II. And used its own law to successfully forecast the cost of aircraft production during World War II. And I'll come back and say a little more about.

So, Wright's law looks like this. If you put it magnetically, it says that X is the cumulative production that the cost of a good or positive technology drops its X to some power omega where omega is technology-specific and omega would be a positive number here for meeting. The cost goes down.

Now, here's an example with four different technologies, transistors global takes Art describes in ethanol where we plot the total cumulative production on the x-axis and the average unit price and dollars on this axis and notice the plots are in log scale. So, the power law turns into a straight way, and this gives you some feeling abruptly. It's not perfect. It varies in how well things stick to these curves depending on the technology. That also seems to be technology specific, but the pattern is pretty clear.

Now you might ask, where does Wright's law come from? So, I would say it's not very clear. But the best explanation is due to James McNerney and me and Redner and Trance. And so, we built on a model actually by John Muth, who, if for those of you who are economists, is the original, the original architect of rational expectations in economics. So, but he also did in those days non-rational thing, like ass ing that engineers just go dart to the dartboard, and what engineers are capable of doing is picking out which dark foes are better than which, and taking going with the best solution they found. And so, it turns out this gives you something like it also actually makes a prediction about how Omega scales with the complexity of the design of the device. As devices become more complex or less modular, then improvement becomes more difficult because you have to coordinate improvements across more pieces.

But let me just say I don't think we fully understand Wright's Law and Wright's law. Certainly, also it's a proxy in that cumulative production is presumably just a proxy for level of effort. And so cumulative production is much easier to measure than level of effort.

We took this, and actually several other laws like Moore's law, and acquired a Library of Technologies, History of Technological Costs through time, and tested this and tested actually several other laws. We tested seven candidate laws, and we did this by, pretending to be in in the past, that, as we pretend to be at a given time in the past. We forecast each future date from that point in the past. If we didn't know anything about the future, we repeated, after all, past days, and we scored the methods based on their forecasting errors, and we made an assumption that the improvement process is the same for all technologies except for parameters.

We'll see more about that in a minute, so that allowed us to test our forecast. Now, the data set we tested on is shown here, and, as you can see immediately. It's a bit of a monthly data set. There were fifty different technologies, chemical processes shown in black here, computer hardware shown in red. Actually, that day we got reward and more of a stop con er goods like beer energy related data sets their genomics.

So, we have several different data sets. The data was acquired by just basically grabbing every data set we could find where we had the cost versus time for a specific technology. And the data was put together by my postdoc, Bayla, Naj. Together with that help of some high school students, we wrote a paper actually testing this testing these different laws, and we showed that basically Wright's law and Moore's law came in with roughly a tie in this data set.

And we also did some things to try and understand how large the errors were, but for various reasons we couldn't properly test this. Now, the side anecdote I can never resist telling is that, well, one of the things we showed was an equivalence between Moore's law and the Wrights Law, and that if costs are dropping exponentially in time and deployments increasing exponentially in time. It's easy to show that Wright's law will haul under those circumstances, and that was the case for these fifty technologies.

This was originally pointed out by a fellow named Sahat Sahel, who wrote several books on technological change in the Eighty’s, and who mysteriously disappeared. One day the hell just vanished, and nobody knows what happened, and Bala became obsessed with Sahal, corresponded with his brother, and then, about two years later, he mysteriously disappeared. So, I’m sorry. I just can't assist telling this. I was, you know Hope Vale is out there doing something good somewhere.

I'm not planning on mystery. So, I'm not obsessed with this. But anyway, thanks to Bailey's work that we managed to put this data set.

Now, how did we model technological change? So, we rewrote Wrights law, which you would write just said, there's a deterministic law that people have historically said, using regressions. We instead turned it into a Time series model. So, we rewrote it as a geometric random walk with drift.

That depends on where the rate of the drift depends on cumulative production of the good. And so, if y now is the logarithm of the cost, so the difference between the amount that the cost drops in, say, a year, is proportional to some constant omega times, the logarithm of the ratio of the accumulative production in here t plus one to your plus a noise term that has some scale.

So, this expression has two free parameters, Omega and S. And we ass e that all technologies follow this process. We take the data, and we fit Omega and pass from the data and on past data. And then we use that to predict future data.

Now, the key thing that we did was to derive expressions. For how accurate the forecasts are that is, let's ass e the world really does follow this process, and to make things even simpler, Let's say that this noise process is an Iv normally distributed.

I thought of the time that was going to be heroic assumption, because I ass e that this would be entail for the structure breaks and so on, but at least for the data step that we applied it to. To our surprise, it fit amazingly well, and we were able to derive expression for the normalized errors. That is, if this epsilon here is the absolute value of the squared error, and this K. Hat. Excuse me, that should have been s in the previous slide. I don't know my slide's consistent, so that's the same as big as there, then, the ratio of these scales like this,

The normalizing constant that determines what the error is, has a term that's proportional, to which is the forecasting horizon end of the future. So, because this is a diffusive process, then there's a there's the squared error goes according to scales, according to Tau, and actually, you know.

And then there's another term that goes according to Tau Square, which has to do with the Estimation error. Where M. Is the number of data points in the data set that you're fitting this okay. So, in other words, you take some data, and you do your best job to fit the parameters.

Then the error you will find a basis expression where this is a student distribution with M one degrees of freedom, so you can just derive this by from saying that the world follows this kind of process. And so, this clock over here shows what happens. The green line is the T distribution. The

The black line is what happens if we just ass e the simple geometric random walk, and the red line is what happened when we ass ed a random walk with put a little bit of correlation in positive correlation. So, in other words, what we found is to make this work. We had to ass e that the improvement rates are correlated from year to year.

And so, this is the result of making six thousand roughly forecasts for all horizons up to twenty years, and then taking all the data from fifty different technologies and twenty different horizons, and a bunch of different forecasts, and putting it all into one bin, and seeing whether it followed this law that we predicted a prior hold, and I should say, if I don't, this is with Francois Hollande.

Well, we've actually showed this first for Moore's law, and then there's a paper with a bunch of authors led by Francois that chose it for my song, because Wrights law was a bit harder to juice. So, what we found is that we could not reject our null hypothesis, which surprised us.

Audience Member: Did you try with opening data and using the datasets to test by making predictions about this?

Doyne Farmer: That's exactly what we did. So, we went. We did this process here. So, we pretended to be in the past, using only the data we had up to that point in time.

And yeah, and we actually did things like systematically look if we only use the five most recent points. We always saw that you always want to use the most data you can. Processes are surprisingly stationary.

Yes, you're treating wars wrong and right. It's all in some sense sort of in people handed way. And there, you know, right conceptually, you're very, very, very different. Yeah, one of them is really the learning by doing deal.

Yeah, if the factory just takes a break, it because of a labor strike for years. One of them is you're marching along the other. See that you're just chilling out for a while. Yeah, I totally agree. And so, this has been debated in the literature which works better. And so, to try and resolve this question Francophone, Diana Greenwald and me.

I wrote a paper where Diana managed to acquire lots of data from World War II, which is a good natural experiment for us, because, unlike these other technologies here, that just, you know, are coming down smoothly over time and are being ramped up exponentially over time in World War II. The Us. Went from the eighteenth-largest military to producing two over three of the military, both sides. By so we got data on the costs of five hundred different types of military equipment. The data was crude. In other respects, we really only had started endpoints.

We did have the whole time series for maybe a hundred and basically what we were able to show by a lot. And let me say the key thing there is that the deployment ramped up enormously, and then it went back down, plunged out as the war was ending, so from that we could see that two things one is, the causality flows from. There is a strong element of causality flowing from cumulative production to cost, that is, and there was also an overall trend. Everything came down during that period.

And so, in other words, there was an exponential trend. But it we ass e the same exponential trend. For all of them there was, but there was also a significant, very significant Wrights law component.

Yeah, we are doing and the exactly right. So that suggests that there is the causality there, and that's why, in what I'm going to show you in a minute. We're going to use Wrights law rather than north long.

No, there’s some problems with that, too, that maybe I can come out and talk about it, you know. Then you get these miraculous things. If you don't continue

Well, I mean you do even better if you have wars law.

So, it's even more miraculous in that case, and a key clarifying remark in the study. We had to war, too. We ass e the same exogenous trend for all five hundred technologies to make this work for individual technologies. You really have to ass e very different trends for different technology.

And so, the exogenous trend that we saw in World War II Doesn't really seem to explain more as well. Here, I think Moore's law is really explained by the fact that lots of technologies happen to have exponentially increasing production exponentially driving across. Let me just emphasize again that different technologies improve at really different rates. Here is coal. Here's the price of the coal in terms of, you know, lots of electricity you can generate what call. And here's soarable political takes. And here's nuclear power.

So, technologies behave really differently. Now we'll move to talking about climate change. How long will it take for technologies to mitigate climate change? What's the right investment strategy which technologies should we support, and how much will it cost?

So let me just show what happens when you apply the kind of thing to extension to four key technologies for dealing with climate change. So, we have in the Solar Vaughan able tanks. We're plotting a logarithm of cost versus the logarithm of an experience. It's just a shorthand for cumulative production. The same for wind and lithe batteries and p two x electrolyzers p two x means power to something, so that you start with electricity. You make hydrogen, and you use hydrogen and make a fuel like ammonia, or one of several other possible lyrics here for forecasts.

So, the Black Doc should the historical values through time. And if you stare at this for a while, you can see that it works pretty Well, that is, things tend to stay near the center of demand that were forecast, and they never actually straight outside of the ninety five percent confidence maybe one point two and you would expect them to do one time and twenty-Two x electrolyzers. You can see our regions are a lot sooner, because we have less data, and the data is noisy. Remember, there's two technology-specific parameters one for the rate of improvement and one for the so this is just to give you a flavor for how this might apply to energy.

I also just want to remark that, you know. Now this isn't so surprising because over the last fifteen years people were very aware that solar energy has dropped to be competitive. But back in two thousand and ten I actually made a forecast nature, and we predicted that it would. Its cost would follow all that of nuclear energy or call. And just to illustrate that at the time, like in two thousand and fourteen. The economist was still saying solar power. It's by far the most expensive way to reduce carbon emissions. So we were, I was moderately successful in forecasting in contrast, the integrated assessment models and the International Energy Agency have done an extremely bad job of forecasting both deployment and cost of renewable energy, and consistently getting pessimistic about it here's a plot where we've taken the data here plotted it in semi-log scale And so you see a rough, noisy, exponential decline, and you see forecasts made in the past that all go out of the slopes that are quite different from the historical slope. Here we made a histogram of about three thousand different predictions by various integrated assessment models that reported in two thousand and fourteen for the cost, for the rate at which solar would drop between two thousand and ten and two thousand and twenty. And this is the histogram of the forecast the most optimistic ones. This is what actually happened.

And similarly, if you look at integrated assessment models, they typically put in what are called floor costs, they ass e that there's some cost where that the technology can't go below. So, in these models they ass e they'll follow Wrights law for a while, but then they'll hit these floors and flatten out. They do this because otherwise, and that the answers become unstable. And this just shows historical floor costs compared with historical costs, showing that it just plunged right through the forecast as they then put up. And my forecast.

These models, by the way, also have deployment constraints. They put in constraints in the rate of deployment, and one of the problems with these

The evaluation of the scenario depends strongly on these two things, which have been more or less arbitrary. We've never seen any evidence for floor costs, so we don't put them in to what we're doing now.

So, we're going to provide an alternative, just doing something very simple. We're going to forecast technological costs conditional on the point. So, we'll throw out some scenarios. We can debate whether they're plausible or not, and we're going to make a few representative scenarios, and we're going to Ass e that sector by sector useful energy demand continues to grow as it has been growing historically at two percent a year. So that is, we ass e that there aren't any dramatic changes in lifestyle or anything like that, apply the methods I just mentioned, and then we just add up the cost of the technologies needed to make the energy sector work.

We say that our fast transitions in our other and key technologies are solar global tanks with batteries and P. And we are picking these well, we're picking these in part, because we think they're good horses to bet on, but also because they have extensive data. There are other things that might may prove to be very important, like just moving power around and transmission lines which we don't put in there because we don't have good historical data, or how the cost will.

So, we use Wrights law to extrapolate, but we extrapolate existing deployment terms in this solution. I'll show you a moment. We phase out fossil fuels over the next twenty-five meters, and we rely heavily on Power Dx fuel for energy storage. So, we ass e that we actually make so much P. Two x fuels that we can run the entire global energy system for a month, even if there's no sunlight or no wind, which is pretty conservative.

We also use pfix fuels for liquid fuels, for heat, shipping and air transport, and usages like that.

Now, in the fast transition, let me actually show you what we do in all three of the scenarios. We investigated these three four key technologies.

So, we shall hear the historical data for the deployment now with solar energy. So, the average, the annual generation in terrawy hours for each year we put a dashed red line through it, just to show the trend. And then this is our fast transition scenario. This is our slow transition scenario, and this is our business as usual.

So, I’m putting this up here hopefully to convince you that we're not doing something that's crazy. We're extrapolating an existing trend. In fact, we begin falling off of that existing trend by the way, as solar reaches maturity, and starts to flatten out. But at this rate it does solar energy and wind. You can see here become dominant after about ten years, and the cut, and completely displays fossil fuels in about twenty years, twenty, five years.

So, this is the same thing for batteries, showing what we ass e for batteries in these three scenarios and p-tax tools, where we ass e for p-text goals that we do see on the existing historical tread for another decade, noting that this is a much less mature technology than the.

And by the way, this is standard and technology literature to ass e that you have s curves. That is, technologies tend to grow exponentially for a long period, and then they flatten out when they reach saturation. So that's what we're doing here. You guys have a question

I'm: a bit of course.

Yeah. Our original business. As usual, we more or less flattened this out to make it a straight line, and the referees complain that we were. We were that that wasn't realistic. So, we saw that our business-as-usual scenario. It didn't matter very much.

They're only price projections for. Yeah. So, we are saying these are assumptions, and we're then going to predict prices. It's true that integrated assessment models try and predict both of those in tandem. I think that's one of their problems. They're doing something they can't really do.

And if you were going to do it well, because these end up, depending very sensitively on the assumptions that I mentioned before through time, which is shown down here for the three different scenarios, and so that when we're doing it for useful energy, which is the energy you actually use, the amount of work that actually comes out vital energy, which is saying one energy would be, How much gasoline we have for our useful energy is, how much, How many joules of work do you actually get out of your car, and this shows the requirements for electricity, for electricity, generation, and storage, and which is then showing that we are talking about ripping the rhythm quite substantially by about a factor of four in order to make this happen.

These show the forecast for some of the key technologies. I should mention that what you can use Wrights, law for oil and gas, and actually because for those for those technologies the omega parameters effectively zero. And so, you just get a random walk if you, if you use Wright's law, Random Walk turns out to be too diffusive. That is, overspend of time longer than about twenty, ten years. Prices just scatter around too much. There's some mean reversion, so we add in for fossil fuels. We probabilistic nature of our forecasts and maybe I didn't stress that enough that that the generalization we make of Wright's law isn't just making point forecast. It's predicting the probability of outcomes at different cost levels at different points in time. And what we tested was that probabilistic prediction. I think these are the one standard deviation and the two standard deviation error bars in the forecast you can see the uncertainty going through time and as we're also comparing to forecast and existing integrated accessible models are making about what the class will be at points of time, just to illustrate that our predictions are still substantially different than those of the other models.

And so, this is kind of the punchline it's showing that if this is under the three scenarios, so fast, transitions flow, transition, no transition. These are we get already right here. Fossil fuel expenditures are coming quite negligible. This is a graph of the media expenditures on energy in each of the years shown so two thousand and twenty to two thousand and seventy, and then looking at the non-fossil fuel expenditures versus fossil fuel expenditures. So, the white is non-fossil fuel expenditures, and the shaded is fossil fuel expenditures.

Here we show the net present cost so we ass e a discount rate, and to put all the different years together. For the investments we ah bring all the costs back to the present by discounting them. And this is what happens if you use a two percent discount rate.

And again, you see the fast transition, the slow transition, and the no transition illustrating the uncertainty, but also showing that the fast transition is on average cheaper than either of the other two cases. This shows the mean cost as a function of a discount rate, illustrating that we don't need to argue about the discount rate the amount we save changes, but the fast transition is the cheapest regardless. And finally, these are mentioned sort of for the vast composition of the time, and compared with the slow transition, or for no transition.

Maybe I’ll just make a remark about technology investment portfolios. We have a paper where we look at investing in technology as a generalized process. And one of the things we showed is that you can't use intuition from finance. One hundred and one mark. Markowitz's portfolio theory doesn't work at all in this domain, because you know it in Markowitz's portfolio theory, says that if you have a bunch of assets, and if you have a positive return on all the assets, if they're not perfectly correlated, then you should diversify by scattering your bets across all the assets.

But if all these assets are following Wright's law, there's something very different, because to make progress down the learning curve you have to invest in them. And so, if you really knew which one had the best characteristics, you should just pick that one in with all your money.

But if you have uncertainty, then you need to spread your resources across the assets. And so, with a situation. You're in a technical technology investment case shows you don't want to put one egg and every basket you don't want to put all your eggs in one basket you want to decide. You want to put Ah, most of your eggs in a small number of baskets, depending on how much uncertainty you have, and what the situation is, and it also shows that you get a lot of extreme sensitivity in these kinds of portfolios.

No, this is. This is a good time you asked about direct to your capture directly or captured with fossil fuels.

Yeah, and that would be additional. So, you would integrate. Put that in with an additional ranch law thing that you have. Yeah, So I can. Of course, we don't have enough data to estimate anything.

Yeah, I mean. There is some historical data on direct air capture, and it doesn't look good. It hasn't come down. The other thing is direct. Air capture depends on fossil fuels being competitive without direct air capture, we predict that we're going to reach a point where that's not going to happen where that will be true fairly soon. So, if fossil fuels aren't competitive without trick or capture, they certainly are competitive with it, and you know also direct. Your capture has all their kind of characteristics that make me think it's not going to come down very much, but that's more of a subjective judgment.

Now, that's different from asking. The question of, are we going to need to use? Direct our capture to just pull C O. Two out of the air in the future? And that's another story. And you know here again what we're predicting. It helps that because we're predicting electricity should become very cheap, particularly if you don't need to deal with storage, and so, so, you know, you could easily imagine that the cost of direct air capture is going to come down quite a bit, using solar electricity.

Now, of course, what I've been talking about so far relies on extrapolating technological trends. And you know, one would like to go to something more fundamental and understanding this problem. And so, we've been thinking about.

Excuse me, different ways of doing that., for one thing, let me just say a little bit about Wright's law. You know Wright's law is a crude approximation at best, because if you pick a given technology, say take solar formal tags. Well, it's composed of different things. There is the cost of building the cell There's a cost of putting up the frame that's going to hold it in place. There is the installation cost, the distribution cost, and these things all behave differently. So generally, you want to decompose things as much as you can, and look at the scaling relationships separately.

This actually relates to a paper another paper of mine with James that James's elite author on where James showed.

If you take industries and you ask what is driving the price index of an industry that most of the improvements in the price. Sixty-five percent of the improvements on average in the price index ministry come from price improvements, and the inputs to that industry.

So, to really do this better. You want to think about the economy on a more fine-grained scale, and you want to think about the interaction of the different technologies with each other. So, we've been working on trying to do that, making more disaggregated models, and maybe just lift the discussion up a little bit. We've been pursuing a method of sort of modeling climate economics in a distributed way with on one that we have our energy systems model, which, as I, you know, Virgin, I just described to you. We're like, for example, making this be the regional instead of law, which already gives significant improvements in terms of providing better advice.

Trying to extract information. We have about technological change. For example, coming from analysis of the patent space where we see interesting patterns like,

Go back and look in the fifties nuclear power. If you do a page rank for Patents nuclear power, it sits in the middle and fossil fuel. Sit in the middle. If you do that now? Renewables in the middle and nuclear power and fossil fuels?

We work on green industrial strategies that is thinking about imports and export stuff that overlaps with the work. Ricardo's group does think about production networks. So, we have models both at the level of industries and of firms going beyond traditional input output of ah models to make dynamic models which we may use up through a comment because it's sort of a sign day. We've been talking about hooking this up with agriculture and land use models more sophisticated models of household. In fact, we have a model of occupational labor, diffusion.

And so, we're doing work. Now we're coupling together our forecast for what's going to happen in the production network when we rewire from fossil fuels to solar power and wind and batteries and electrolyzers.

And then how does that play out in terms of occupational labor, and then thinking also about treaties and government, and how that interacts with all of this. So, we're basically pursuing independent projects each of these pains, but also starting to think about how to paste them together, at least pairwise or triple-wise to make improved forecasts, and put in more fundamental information.

I think I think this suggests that doing the things that we need to do to support this.

Yeah. So well, that's an interesting question in that. When you go back and stop start looking at raw materials like things, your things you mine out of the ground. It's a quite striking, not well appreciated fact that actually James was the first person to introduce me to. James went, looked at the Usages data, and observed that that century of data for more than one hundred different and ah, he fit some time series models to those, and what you see is once you adjust for inflation, They're all flat except one exception. Ah, industrial diamonds! Now, industrial diamonds aren't something You might out of the ground they've dropped by a factor of one hundred. Everything else. Nothing has changed by more than a factor of ten in price over a century.

Now that suggests what so raw materials? If there are things you're mining out of the ground. Why? Because they aren't going to get much cheaper. What gets cheaper are the processes, for you know making them into things, and that's where That's where the progress

I mean. The other thing you might have asked about is shortages of essential materials.

My impression, Yeah, there may be some shortages of essential materials, but when you go back through the history of technologies, shortages of essential materials. Whenever there's a bottle of actuative essential material. Another material is found that provides a workaround, and there are quite a few historical examples of that chlorophyll carbons for refrigeration, and many others.

So, to wrap up, I think. What are we shown? Well, first of all, I think it's important it we've been. It's been clear for some time that renewables were getting cheaper and cheaper, but people were still viewing the renewable energy transition or the green energy transition getting rid of greenhouse gases as a burden. Who's going to take on that we're saying, Actually, no, it's going to be cheaper than going on the way we are now. So even if you're a climate denier, you should be willing to j p on board to make this happen. And it also, I think, reframes the whole discussion from having climate change be a burden to have it being an economic opportunity, and the players who get involved. The firms and the countries who get involved are likely to require the expertise to be players in the economy in the future. And so, I think it really reframes the whole discussion.

Now we also, you know, as I said, we were just throwing out the scenarios as assumptions, not as something we're driving from first principles, so we try to make them plausible by extrapolating from existing trends.! But stuff has to happen to stay on those trend line in particular, I think the biggest one is the grid. We need to expand the grid globally by a factor of four, and that's a pretty big change. It's expensive. But if you look at paying ah, you know a grid even. Let's say the grid cost was five hundred billion per year. I think that's higher than the number we'd settled on now, but because we're already paying four trillion a year for energy.

So, if we can reduce energy costs by, say, a factor of two. Then that dwarves the cost of building out the grid. But we have to worry about the political roadblocks to building out grids and doing that stuff. My son works for the Federal Energy Regulation Commission, and he said that there's enough proposed renewable energy projects.

Sitting in the pipeline. We not approved yet, but having been given to the Commission, if they approved them all, we would more than double the electrical capacity in the Us. They can't approve them all, because there isn't a good capacity to put them all on. So, we really have to deal with that

We did look at stranded asset issues, by the way, and we found that that's not so bad because you replace a gas station every twenty-five years.

We're turning over our infrastructure all the time. So, we have to do is let the old infrastructure in the spot expire and replace it with infrastructure for renewables, and we, we pretty much get there without spending that many assets we do need to push on storage technologies. We need to stay on that trend for twenty years if we're going to make it. And. And, as already said, we reframe climate change as an opportunity. And you know I found myself. I was sailing along the North African coast this s mer. So, this is a photo I took from my phone call. This would have been in Morocco, and, you know, dealing with the bizarre police states in Algeria that have to do with you, have a military government that controls the key resource oil and keeps the whole country under an iron fist.

Oh, I think there are going to be substantial geopolitical changes as a result of this transition. Twenty years is pretty fast, and countries like this are going to have serious problems dealing. So, I have a question, you know, but some large-scale Princeton study of a dead zero. Economy said, you have four and X: Yeah, that's what we're saying to it. And the course of keeping that I think it’s important to remember, though, that today in Massachusetts, I suspect in Brooklyn it's not. A better generation is about forty five percent of our electricity. It includes some storage and some,

or back on capacity, or whatever you want to do, a factor in the overall price.

So that makes me wonder whether you just because solar and wind get super cheap balancing, it is maintained. And then it's sort of like, well question, you know, module cost versus balance of systems.

Yeah, but burning cards are harder to see on the transmission, And the same is true for hydrogen. And right now, the electromagneters are about twenty percent of the cost of one thousand one hundred now so well, and the other cost, and you know the stuff where learning birds are. They happen? But that is obvious that giant storage tanks are going to come down and roast around, and that's not even including the distribution of that.

And so, I just. I just wonder what you think about that tension. Yeah. Well, so we you know, we agree that the grid is going to be expensive, and we ass e we spend a lot of time researching. You know what the literature says about what it cost to install red new grid capacity, and there's of course, a difference of whether you're talking about transmission or distribution, and there's a difference of whether you're putting more wires in an existing line, or when you have to build a new line,

but we made pretty conservative assumptions. We ass e there's no learning that the costs remain the same as they are now. We just scale them up because we're doing a factor of four more. So, we increase the cost by a factor of four or five, what I think the people you're talking to Don't realize is if the other cost comes down.

Then, you know I mean, you know, if you can, you can get an upper bound by just asking What if the generation were free? Right? So yeah. So, we try to do that now, whether I mean we're very open to seeing whether our numbers are, you know, having people go over our numbers and see whether they are correct or not. So, I agree. The devil's in those details.

It's also true that you have to. It's very tricky. You see, when you get to fossil fuels, there's no meaningful statement about cost of fossil fuel production, because you have places like Saudi Arabia, where it's, you know, a few dollars a barrel, and shell oil in Canada, where it's fifty dollars an hour, and so the price of fossil fuels depends on what the marginal cost is for whatever the marginal supplier is at that time, and that's why it's so valid.

That's why I think one of the big side benefits of renewables is that we're going to see much less volatile energy costs as a result.

Because there's no reason to believe we're going to have that kind of mechanism where we have such a radically different set of costs. You're moving up and down the marginal price point all the time based on fluctuations in supply and demand that curves going to be a lot. Cloud.

But yeah, sorry I’m trying to remember. I'm killing. I did. Did I address all of the things you brought up a pretty small percentage of the concentrated.

Well, of course, you know, in the in that plot I showed with the different technologies, there are a bunch of lack of chemical processes, and one of the things you see is chemical processes. All tend to behave about the same way. So, it's one of the more understood classes of technologies, and there are big economies of scale. From things like tank size, we get surface vol e, relation, surface, vol e relationships help.

So, I think electrolyzers will come, I mean, from looking at related technologies. I'll be very surprised if they don't come down in price as roughly, we would expect from the industry we have so far.

But time will tell. So, she thanks for everything That's another way to answer Dan's question in that, if you know solar, if you become so cheap you can just paint it on the roof of a building, and then the building is collecting solar energy.

Then you have a lot of possibilities for decentralized generation. You still are going to need to deal with storage

now. Electric vehicles already are going to increasingly provide an example of decentralized storage, and the costs we were looking at were caused for

ah industrial installations. There is a shift for a localized, ah decentralized generation, but if the costs come down enough, then that shift won't matter much.

I would like to understand connection between different technologies. Evolution, of course. Or

Yeah, you really have to go technology by technology. Now, there is a question of what is a technology? you know, there's twenty or thirty different flavors of solar photovoltaics, and it's not clear if all of those follow the same scaling, many of them likely to, because there's somewhat similar processes.

You know, a technology is always evolving through time. So, the technology is always changing.

Nonetheless, what the data suggests is, you can sort of designate something as a technology and look at its costs through time. Now, coal-fire electricity is a good example where up until the eighties people didn't care too much about sulfur emissions, and then people started caring about it.

So, you might think a joule of energy is a joule of energy. But there's dirty jewels and clean jewels. There's intermittent jewels and steady jewels. So, there's dangerous jewels and Saint Joules. So nuclear power suffered from that problem. So, your designation of the technology can shift, and the cost can shift if the requirements are placing on it, change as they did it with coal and nuclear power.

But this goes back to what I was alluding to with this plot I showed back here for our attempt to try and take a more holistic view of the economy, and use that to think about climate economics, because we really, we would like to be able to track all the technologies comprehensively and think about the ecology of interacting technologies and use that as the framework in which we think about things, rather than using what might be an inappropriate level of aggregation. And that puts noise on what's happening, because you expect things like distribution scale very differently than technology parts.

So, we're increasingly moving towards trying to find the right levels of decomposition and then combine things, often splitting things into their inputs. In the paper James wrote a paper on coal. Ah, in I don't know. When was it? Two thousand and thirteen or fourteen back there somewhere, and what two thousand and eleven where, you know we decompose coal-fired electricity into operating and maintenance. The cost of building a plant, and we saw that those three things really became different and useful to split that.

So, it's a tricky question that requires a lot of case-by-case judgment.

I don't even trip from the very back a lot of the investment in these technologies. There's been, you know, public investments variable of refinement, and we found that that investment makes a difference as the rate of change.

Yeah, so it's a bit hard to say in a rigorous way. It certainly seemed like there's in Germany. There were massive subsidies for solar energy that did correspond to a significant drop in price. This is usually called feed in Terrace. And when you think about when you actually go back, if you think about these curves, I’m showing here. Okay, you know, how did solar get solar is very expensive back at this point in time, and its cost keeps dropping well the very first toll ourselves in one thousand nine hundred and fifty-eight, Bangladesh.

Then people started using it, for you know weather stations at the South Pole places where it cost wasn't a big issue, and then it got more and more in ex uses, so that the essential thing for a technology to roll out like this is that there be a set of niches such that it can keep advancing along, and as it does, its costs can drop, and so it's important to keep that going, and I think the cost supports the price supports were important in keeping that happening.

But I Don't haven't seen a good study that really teases that out, and shows cause and effect, and it's probably pretty hard to innovation is tell, you know that's critical, and you know there’s Marian I'm, at Matsicado. But others have done studies, too, showing that public investment and technology has played a major role in seeding these things.

You're essentially paying for the material and, the input energy in the house of when you,

I think we will evolve in the new technologies. We don't we're agnostic about that. But what's predictable is the future costs what is not predictable, and the solutions that deliver those costs. I once heard of fascinating talk by a guy who was a chip designer. This was at a meeting in Sfi maybe fifteen years ago, and they were talking about Moore's law. So, you guys all ass e Moore's law is just inexorable, he just, and he said, having been a chip designer that wasn't the way it felt to us at all. We felt like we were looking at a brick wall.

We couldn't see any way to get to the next step, and then somebody would come up with an idea, and we would break through and find a solution. And miraculously that happened over and over again, and Moore's law has sustained through that whole period.

You know. Some people say Moore's law is just a self-fulfilling prophecy. That's clearly not true, because, for one thing, when Moore wrote the law down. Nobody else was aware of it, and it is true that it has been useful for people to plan right, and I mean a great example of this is my friend Emily Ray Smith, who was in better of Ah CGI graphic techniques for Pixar. And he said they basically had it already to go five years before it actually got rolled out.

They didn't know they just had to wait for Moore's law. They had all the techniques down, and They waited five years, and sure enough then, and then then toy story was the result.

So, it does allow you to plan, but it's not like it's a done deal, or that it makes itself happen

But if the technology is predicted using silicon data that you were saying to keep it going?

The prediction can come from any.

And you know, you see this if you look at, say information technologies where we made a transition from mechanical information devices to vacumtubes, to discrete transistors, to integrated surface.

So, each of them has its own curve, and but the envelope that they followed is quite smooth, despite the fact that you're leaping from technology and technology.

Why do you think we've here? It hasn't it the grades more as well, because this also involves the technological.

Yeah. So, there's It's a mystery. I don't think we have a good understanding. Now people have speculated that certain properties like modularity matter. Actually, our theory with James has something like modular, and we called design complexity.

And if you go out and test that you do see that? And there' been a couple of tasks from our theory. Ah, that indeed, as the as the design complexity goes up, the learning rates tend to get worse.

But it's very noisy, very noisy, and it's like one of the facts that I think is amazingly strong. Is this one about stuff you mine out of the ground.

But you know, if you look at the way you extract oil now, it's radically. It's incredibly sophisticated compared to what was done a century ago.

Now there's been there. A trade-off between the whale gets harder to find because we pulled so much out. Technology gets better to find it. But it seems very surprising that for one hundred different minerals you discover that those two trade-offs are just balancing each other across the century, you'd expect U-shaped curves that it would drop, and then it would go back up, and that's not what you. See? So, there's something else going on there, and you know, nuclear reactors are.

Well, a couple of things. If you actually, if you read an account of what they have to do to clean up a spill in a nuclear reactor. It's pretty complicated.

And so, they're complicated pieces of technology. This the French have maybe had a little bit of success in standardizing it so to make a cookie cutter. But it Hasn't really paid off. French nuclear power is very expensive, too.

There's a school of people very strongly advocating for making modular nuclear reactors. Small nuclear reactors not going to fly, because they're fighting a big scaling law. To begin with, there's a reason why they made them big to start with, and you have to make that back up, and there's just no way they're going to do that. But! But there's a mystery. There's still a mystery. There's no way to predict a priori which technologies are going to improve quickly. And which are it's a big problem. It’s fascinating problem.

Yep, Sorry, he hadn't said. Had a chance Yet it might be that it's general. Yeah, I know. I see your arg ent now, and There's certainly an element of truth there at the same time, you know. If you look at, say the Danish experience with windmills. The Danes were early adopters of windmills and acquired expertise in windmills, and then they ended up becoming designers of windmills and have an industry around one else. As a result of that, this goes back to what Ricardo always stresses about dawn.

So, you get the know-how by being there early on and if the technology.

And so that's the flip side of the argument for the concern that you raised.

Well, so we did a study of geographical variation in solar and wind prices, and a couple of things strike you. First of all, you might think that countries that have good solar and wind resources would be the one to install the most it's roughly uncorrelated. It really has to do with politics, and there is a significant geographical variation due to things like how often the sun shines and institutional factors about installation costs in country A versus country. But, roughly speaking, if you look at solar energy the ins in the course of a decade at a ten percent per year improvement rate, you move from the ninety fifth percentile to the fifth percentile in the course of a decade. In other words: If you're the place to have the most expensive energy in the ninety fifth percentile, then by a decade later, you'll have the same energy cost that the people that were in the fifth percentile had,

So that roughly gives a scale of the geographic variation versus the temporal variation that said, we are trying to build out a regionalized version of our model to deal with that kind of thing.

Ricardo. Yes, and we said it was correlated with the information rates, or and you know the sunshine for free,

you know.

No. So please the stuff we're going to do a variable cost right? So, when you work out the cost, the interest. Yeah,

Our purchase agreements out there,

you know, twenty dollars, and they have one hour soldier,

you know. You know the and it doesn't. Generally, it could be that we didn't look at interest rate. So that might be. =

No, I don't yield.

It's all right.

Yeah, one of the interesting things that people didn't anticipate. That's boosted solar.

The rollout has been that people have been willing to offer very low interest rates for solar projects, just because, in contrast a to nuclear power projects, they've been very reliable in terms of. If the project gets going, it tends to be delivered. The uncertainties are much smaller than ah in other domains, so Solar has benefited. Certainly, it's benefited from the fact that we've had low interest rates now for two decades,

but it has among energy. It gets low interest rates because it's reliable with us.

It's substantial, and you know, I haven't looked into that in as much detail as I would like, but the studies I've seen Don't suggest that it's, and they suggest it's doable, and that the you know there was a review article in Pts. Six eight years ago, comparing the environmental impact of different technologies and solar, even when you take into account the need for rarer minerals and so on. It still gets the best rating relative to the competitors.

But I'm not thinking of investing in my yeah, you, okay? Well, yeah, well, you know, and I think that, by the way, I think part of the reason mining has this property of being so volatile, is because it requires such long-range investments. I grew up in a mining town, and you know, when the price of copper goes up, everybody goes to work, and when the price of copper goes down, they get laid off, but it takes five years to start up even to start up. An existing wine takes years of preparation.

So those big lead lag effects.

Yeah, one question about the fifty-six of investments,

You know the everybody is going to the cable for this

Well, yeah, so we're not arguing there. There might not be bottlenecks that would not this top half. I think One thing that you know that I reflected on in doing this is that we're very bad at thinking about exponential rates change in their mark. So, extrapolate this roll out. You hit the other curves up there.

So, we're really at a critical point. Right now, we're you know, solar energy is still just a percent or something of global power, but it's about to burst forward on the stage, and then the question is, how far can it go before it starts any?

I mean my guess is that this won't be a It'll be a little b p, but it won't be a long-term change in the trend. You know one of the reasons why I think renewables are going to have one of the positive side effects or energy security, which, with Ukraine, we see, is pretty important. It means, if you know, maybe, that if you're in England, you have to pay more for your power than you do if you're in Arizona. But if you're willing to pay for it. You can get energy security, regardless of the natural resources you have regarding as we do now.

So, I think that's going to be a big factor of pushing countries toward it. And so, we may even benefit from this isolationism in that regard, because it's yet another incentive to go renewable.

Well, so one thing to realize is that renewables come in third when it comes to subsidies. So, fossil fuels get the most subsidies nuclear power against the second, and then and then renewables, even adding them all up. Yeah, so to do this, I mean, when I was talking to my side, I mentioned where it's. The Federal Regulation Commission said promptly. Don't quote me on this.

Thank you. Good set of questions. I really enjoyed the discussion. Thank you.