1
00:00:12,860 --> 00:00:21,890
Hello, everyone. My name is Charles Gun-Free, I'm director of the Oxford Martin School, and let me welcome you to the Oxford Martin School.

2
00:00:21,890 --> 00:00:27,260
When I buy a book, I scribble in the book the date that I bought it and where I bought it.

3
00:00:27,260 --> 00:00:32,750
So I know where I first became aware of aware of and former speaker.

4
00:00:32,750 --> 00:00:35,810
And it was, and I've just looked it up, saw myself down below.

5
00:00:35,810 --> 00:00:45,320
It's July 1998 and I was in Seattle at the time and the book was James Gleeks book on Chaos.

6
00:00:45,320 --> 00:00:54,830
And it has a whole chapter on the dynamical systems collective Santacruz, of which Darwin was one of the one of the members.

7
00:00:54,830 --> 00:00:57,860
This was an extraordinary group of people in the late seventies.

8
00:00:57,860 --> 00:01:06,860
And you all before you eyes then weren't you who it just expanding work and bringing forward the field of complexity science?

9
00:01:06,860 --> 00:01:12,620
I have to say I find it deeply depressing because I was just doing just out of my Ph.D. myself,

10
00:01:12,620 --> 00:01:16,820
and no one was going to write a chapter on what I had done in my day.

11
00:01:16,820 --> 00:01:25,820
So that was one of the founders of the field of complexity science and has gone on to do very many major things at Los Alamos.

12
00:01:25,820 --> 00:01:35,360
And then most recently in Oxford, which he joined nearly 10 years ago and then today is the co-director of the Oxford Martin programme on Post Carbon

13
00:01:35,360 --> 00:01:42,980
Transition and a principal investigator in the Oxford Martin programme on 10 Technological and economic change.

14
00:01:42,980 --> 00:01:51,830
He's also a professor in the Mathematical Mathematics Institute here in Oxford and external professor at the famous Santa Fe Institute.

15
00:01:51,830 --> 00:01:57,470
And then going to be talking this evening on how complexity can resolve the crisis in economics.

16
00:01:57,470 --> 00:02:07,040
Then. So thanks, everybody for coming.

17
00:02:07,040 --> 00:02:10,640
When I was invited to give a lecture on an evolving economic thought,

18
00:02:10,640 --> 00:02:16,220
I couldn't resist trying to impart what I've learnt in my twenty five years

19
00:02:16,220 --> 00:02:20,960
or so of working in economics as a physicist who sits in a maths department,

20
00:02:20,960 --> 00:02:24,930
who does economics, who's also dabbled in biology and a few other areas.

21
00:02:24,930 --> 00:02:32,600
And and so I'm trying to to sort of distil the main things I've learnt during those twenty five years.

22
00:02:32,600 --> 00:02:35,930
This is also, by the way, sort of a I'm writing a popular book.

23
00:02:35,930 --> 00:02:42,290
And so this is the first test run to try and present what I'm trying to say in the book.

24
00:02:42,290 --> 00:02:48,440
So economic thought changes through time. You know, when we're in high school,

25
00:02:48,440 --> 00:02:55,670
we're taught the scientific method and the retarded as though it's just cast in stone and everybody does it that way.

26
00:02:55,670 --> 00:02:58,070
But in my experience, that's not at all true.

27
00:02:58,070 --> 00:03:04,880
It's vastly different in different disciplines, and it evolves through time and economics is evolving through time.

28
00:03:04,880 --> 00:03:10,280
It's becoming more empirical science. I've really seen that in the last decade,

29
00:03:10,280 --> 00:03:20,120
but it's experiencing tension because the changes it's finding are not completely compatible with the theoretical framework that's been set up.

30
00:03:20,120 --> 00:03:26,930
It's also had some problems where there have been voices of protest that the 2008 financial crisis,

31
00:03:26,930 --> 00:03:32,960
where the models were completely abandoned during the crisis because it was clear they weren't relevant to what was going on,

32
00:03:32,960 --> 00:03:39,110
and there were a lot of criticism that they should have done a better job of being ready for something like that.

33
00:03:39,110 --> 00:03:49,550
Inequality, actually, we were at the OECD a few weeks ago, and Gabriela Ramos said that bad economics led to bad politics.

34
00:03:49,550 --> 00:03:57,110
So we're experiencing, you know, a period where people are sceptical about experts because they feel like they got

35
00:03:57,110 --> 00:04:01,820
bad advice from them or the experts in many areas don't agree with each other,

36
00:04:01,820 --> 00:04:11,180
which never looks good. And and we're, you know, we're experiencing the consequences of layers of society.

37
00:04:11,180 --> 00:04:15,500
We never thought we were going to see them from are significantly caused by bad

38
00:04:15,500 --> 00:04:20,750
economic models and bad economic policies being put into practise as a result.

39
00:04:20,750 --> 00:04:27,170
And then the big one is climate change, which is, after all, caused by the economy, and we need to figure out how to fix it.

40
00:04:27,170 --> 00:04:33,620
So we really need guidance from economics, and I would argue we're not getting the guidance we need.

41
00:04:33,620 --> 00:04:38,720
So this is these are these are the factors that led me to say that there's a crisis.

42
00:04:38,720 --> 00:04:40,460
But even more specifically,

43
00:04:40,460 --> 00:04:47,480
there is there is a cognitive dissonance within economics because there's a theoretical machinery that's based on rational expectations.

44
00:04:47,480 --> 00:04:51,200
Behavioural economics says this is wrong. OK.

45
00:04:51,200 --> 00:04:56,450
You may not know what this means. I'll explain in a bit. Can adding friction save it or a bigger change is needed?

46
00:04:56,450 --> 00:05:02,930
So I'm going to try and discuss this during the course of my talk. I'll come back and explain what these words mean now.

47
00:05:02,930 --> 00:05:03,710
But before I do,

48
00:05:03,710 --> 00:05:10,910
I want to talk a little bit about what the economy is because I think if we if we if we think the right way about what the economy is,

49
00:05:10,910 --> 00:05:14,960
then we set the scope and the goals of what we're trying to do differently.

50
00:05:14,960 --> 00:05:22,370
And that affects the way we talk about things. Now I'm going to argue that if I had to describe the economy in three words,

51
00:05:22,370 --> 00:05:29,900
it's the metabolism of civilisation and that it converts natural resources and human effort into goods and services.

52
00:05:29,900 --> 00:05:37,250
It provides us with things we need and don't need, but it provides us with all the stuff that processes a lot of information to make that happen.

53
00:05:37,250 --> 00:05:42,890
And it's very much like a metabolism in a biological organism, which takes some form of food,

54
00:05:42,890 --> 00:05:47,930
breaks it down into different components, and then reassembled it in the stuff that we need.

55
00:05:47,930 --> 00:05:51,980
That's just what the economy is doing. The difference is, instead of chemicals,

56
00:05:51,980 --> 00:05:58,340
it's happening in a much larger scale level with material stuff and that the

57
00:05:58,340 --> 00:06:02,420
economy is a truly remarkable thing that I don't think we properly appreciate.

58
00:06:02,420 --> 00:06:10,100
It coordinates, and I will argue it coordinates and amplifies the activities of ecologies of specialists.

59
00:06:10,100 --> 00:06:14,750
Now what do I mean to that? Well, we're all specialists. OK. I am a physicist.

60
00:06:14,750 --> 00:06:16,090
I'm working in economics.

61
00:06:16,090 --> 00:06:21,560
I'm sure if we went around the room, we'd see lots of different specialities and what people are doing and you go more broadly.

62
00:06:21,560 --> 00:06:29,270
I mean, there's there's tens of thousands of occupations and skills, and we all know how to do a specific thing.

63
00:06:29,270 --> 00:06:36,530
But we're all interacting with each other, just like, say, in a biological ecology animal specialise in doing a certain kind of activity.

64
00:06:36,530 --> 00:06:47,470
And then ecology is about the study of how those interactions happen in a more holistic way and how those interactions take place.

65
00:06:47,470 --> 00:06:53,500
It allows us to do things that we could never do on our own. Imagine that each of us had our little plot of land.

66
00:06:53,500 --> 00:06:59,650
We divided the Earth up. We all got our little plot of land. The rule was You can interact with anybody.

67
00:06:59,650 --> 00:07:06,490
You've got to survive. How many of us would survive? Probably close to none.

68
00:07:06,490 --> 00:07:13,390
You know, we owe our lives to the planet. Could never support the population has not by a factor of a thousand or 10000.

69
00:07:13,390 --> 00:07:19,000
Even if we didn't have the economy to boost what we're doing now,

70
00:07:19,000 --> 00:07:26,560
just to give a little spot a view of richness of just one thing take the laptop that I have sitting here in front of me.

71
00:07:26,560 --> 00:07:35,120
Here is the physical supply chain of a laptop is constructed by some fellow named Leo, and a lot of hard work went into this.

72
00:07:35,120 --> 00:07:41,770
And you can see with the little numbers and the arrows showing how the basic raw materials get taken out of the ground.

73
00:07:41,770 --> 00:07:48,460
They get shipped around the world to other places and processed. They get smelted assembled into larger parts.

74
00:07:48,460 --> 00:07:55,450
Those parts could ship to new places. Those get assembled a laptop, get shipped somewhere, you know, and the tax bill goes to Ireland.

75
00:07:55,450 --> 00:08:00,160
And so. So it's a remarkably complex thing.

76
00:08:00,160 --> 00:08:05,350
And if you think about it, compare it, say to a human birth, which is a remarkable process.

77
00:08:05,350 --> 00:08:12,130
I mean, the closest thing we have to a miracle. But it does all happen inside of a fairly small place takes place in about nine months.

78
00:08:12,130 --> 00:08:16,870
This takes a couple of years from the stuff getting dug out of the ground, the laptop getting assembled,

79
00:08:16,870 --> 00:08:21,550
and it's coming from all over the world and miraculously coming together in one place.

80
00:08:21,550 --> 00:08:25,540
Now imagine we had a Google economics.

81
00:08:25,540 --> 00:08:30,850
So what do I mean by that that instead of just having, you know, map you could like,

82
00:08:30,850 --> 00:08:37,990
take your phone and you can point pointed at a building and it would tell you about the economic activity that was going on there.

83
00:08:37,990 --> 00:08:41,330
Imagine that you could see the flow of the goods and services in and out of the building.

84
00:08:41,330 --> 00:08:46,690
So, you know, if it was the Apple headquarters, you'd see some some part of this picture.

85
00:08:46,690 --> 00:08:51,670
You can see the stocks and flows of capital because of course, to make all this stuff,

86
00:08:51,670 --> 00:08:57,100
you've got to have money flowing in because money is like 8p in biology.

87
00:08:57,100 --> 00:09:01,390
It's the energising thing that makes stuff happen.

88
00:09:01,390 --> 00:09:07,660
People that have money can just wave their wand and people will start working and doing stuff.

89
00:09:07,660 --> 00:09:11,260
There's a web of contracts that is to produce all this stuff.

90
00:09:11,260 --> 00:09:15,100
All of these providers have contracts for their workers, have contracts with each other.

91
00:09:15,100 --> 00:09:23,770
These contracts link their activities together and and tell us, interact through time because contracts have conditions.

92
00:09:23,770 --> 00:09:30,040
And if this happens, then that happens. There's demography and occupational capabilities.

93
00:09:30,040 --> 00:09:36,010
There's a heterogeneous population on the globe that's doing very different activities to make this happen.

94
00:09:36,010 --> 00:09:39,370
It tells us it would tell us about wealth and poverty. Who's getting rich?

95
00:09:39,370 --> 00:09:44,980
Who's getting poorer? Tell us about ownership. Well, we know it's very concentrated.

96
00:09:44,980 --> 00:09:52,660
It could tell us about the ecology of innovation. Who are the thinkers that are that are influencing the people that are doing this stuff?

97
00:09:52,660 --> 00:10:02,170
It would tell us about the physical and environmental impacts, how much carbon is getting emitted, how poorly are they doing at recycling materials?

98
00:10:02,170 --> 00:10:07,210
It would tell us about the regulatory constraints that are shaping what's going on, which is also an essential part.

99
00:10:07,210 --> 00:10:13,030
And I put Typekit because this is just the beginning of the story. Now, how complicated is it?

100
00:10:13,030 --> 00:10:17,560
Or at least let's put some numbers on how big it is. Well, the production network.

101
00:10:17,560 --> 00:10:26,320
So if we really look at this for all the production of everything on the globe, we're talking about the order of 50 50 million firms.

102
00:10:26,320 --> 00:10:34,210
Those firms have billions of physical links to each other because each firm, as we see in this picture, is getting inputs from the other firm.

103
00:10:34,210 --> 00:10:41,770
So each firm is profiting from the innovations the other firms making the whole thing's very closely linked together.

104
00:10:41,770 --> 00:10:48,520
There's there's a network of households, the order of two billion of them, the order of three and a half million workers.

105
00:10:48,520 --> 00:10:53,680
There's trillions of links to the things they consumed, the people they work for, the contracts they make.

106
00:10:53,680 --> 00:11:00,130
And there's a web of contracts. There's probably the order of a trillion contracts. They're active on the Earth at this point in time.

107
00:11:00,130 --> 00:11:03,940
And it's a dynamic contract expires. A new contract gets made.

108
00:11:03,940 --> 00:11:07,600
It might be different, OK, the economies heterogeneous.

109
00:11:07,600 --> 00:11:15,790
So this is a picture of how prices have changed for various goods in the United States over the last 20 years.

110
00:11:15,790 --> 00:11:24,760
So at the top, you look at hospital services, which have gone up by a factor of two hundred and twenty five percent or so in the US during 20 years.

111
00:11:24,760 --> 00:11:29,110
And you look at the bottom and television sets that have dropped in price so much that it's

112
00:11:29,110 --> 00:11:33,550
actually hard to tell just how close they were getting to being free from this picture.

113
00:11:33,550 --> 00:11:41,740
And there's a lot of stuff in between. So it's a very heterogeneous process. Take.

114
00:11:41,740 --> 00:11:47,830
So let me just say so if you read in the newspaper that inflation is two per.

115
00:11:47,830 --> 00:11:58,990
Well, what does that really mean? That means that for hospitals, it's a whole lot more than two percent, and for televisions, it's minus 15 percent.

116
00:11:58,990 --> 00:12:07,810
So it's a very heterogeneous number, and the average inflation actually tells you only a very crude thing about what's actually going on.

117
00:12:07,810 --> 00:12:11,590
Take another figure. Like unemployment, you read that unemployment is five percent.

118
00:12:11,590 --> 00:12:21,350
What does that mean? Well, here's a histogram down there of in the lower left corner of unemployment rates for different occupations.

119
00:12:21,350 --> 00:12:26,560
So if you divide the world into, say, four hundred and fifty or so occupations, what are the unemployment rates?

120
00:12:26,560 --> 00:12:30,100
Well, if you're a nurse, the unemployment rate's really low.

121
00:12:30,100 --> 00:12:36,820
And if you're an actor or a boilermaker, the unemployment rate's, you know, twenty five percent or more.

122
00:12:36,820 --> 00:12:43,750
And the network picture that we've constructed over there on on the right is showing how if you

123
00:12:43,750 --> 00:12:49,990
organise occupations based on similarity of their tasks to each other and or organise them,

124
00:12:49,990 --> 00:12:54,490
alternatively, we could have organised on job transition probabilities between occupations.

125
00:12:54,490 --> 00:12:56,050
Then you see, there's a very clear structure.

126
00:12:56,050 --> 00:13:02,180
There's some occupations down in the construction industry in the lower left of the picture with very high unemployment rates.

127
00:13:02,180 --> 00:13:08,110
There's occupations in the health industry and the right with very low unemployment rates,

128
00:13:08,110 --> 00:13:12,490
technological progress, essential thing driving economic growth.

129
00:13:12,490 --> 00:13:21,850
If you make a histogram of of of the rate of improvement of different products, you see, it's again extremely heterogeneous and very skewed.

130
00:13:21,850 --> 00:13:24,220
That is a kind of a bell curve in the middle.

131
00:13:24,220 --> 00:13:32,650
And you can see if you look at the x axis, which is showing the average annual growth rate over a close to fifty year period.

132
00:13:32,650 --> 00:13:36,400
Then you're seeing that some technologies are.

133
00:13:36,400 --> 00:13:43,960
Most of them are in the middle of this bell curve. But there are some technologies out on the tail like things relating to computers that have grown

134
00:13:43,960 --> 00:13:49,990
at much faster rates than and have improved at much faster rates than the rest of the economy.

135
00:13:49,990 --> 00:13:56,680
And if you guided even to a more fine grained level, and here are the pictures changed, it's a little you have to.

136
00:13:56,680 --> 00:14:01,510
We're representing things that are in a different way because the y axis is now the improvement rate.

137
00:14:01,510 --> 00:14:08,500
So you start on the left side with optical telecom, which is improving at an astounding sixty five percent per year.

138
00:14:08,500 --> 00:14:15,550
And on the other side, you've got a more, let's say, basic technology like milling machine or electric motor,

139
00:14:15,550 --> 00:14:22,000
things that have been around now for hundreds of years that are improving at rates of a few percent per year.

140
00:14:22,000 --> 00:14:26,800
So there's a huge. Disparity and heterogeneity and these improvement rates.

141
00:14:26,800 --> 00:14:35,270
Now finally, people are heterogeneous, and we have to remember when we think about the economy, we're all different and we think differently.

142
00:14:35,270 --> 00:14:44,200
We have different needs. And that's another important factor that needs to be taken into account whenever we think about economics.

143
00:14:44,200 --> 00:14:48,100
Now there's a caricature of economics as being a counting plus behaviour.

144
00:14:48,100 --> 00:14:53,200
This leads some things out. Maybe come back and say a bit about it, but let's just start there.

145
00:14:53,200 --> 00:14:59,350
And so the accounting is, in a sense, the easy part because of accountings.

146
00:14:59,350 --> 00:15:06,850
Accounting is mechanical. And you know, every household, at least conceptually has a balance sheet.

147
00:15:06,850 --> 00:15:12,100
And on that balance sheet, as the assets and liabilities, the things they own, the things they owe to other people.

148
00:15:12,100 --> 00:15:20,620
And as I already said, there's the order of two billion balance sheets in the world and there are thousands of entries,

149
00:15:20,620 --> 00:15:27,160
I would guess, on the average balance sheet. Certainly, people in the developed world have thousands of entries on our balance sheets,

150
00:15:27,160 --> 00:15:32,230
and that means that when you add everything up, we have, you know,

151
00:15:32,230 --> 00:15:33,070
what's already said,

152
00:15:33,070 --> 00:15:41,950
there's trillions of contracts and our balance sheets are interlinked because my liabilities or somebody else's assets in general.

153
00:15:41,950 --> 00:15:48,400
So every time you have an asset liability relationship, those balance sheets get linked together every time you have a contract.

154
00:15:48,400 --> 00:15:55,570
Those balance sheets are linked together. So you have this vast network with two billion links and lots of with two billion nodes,

155
00:15:55,570 --> 00:16:02,260
trillions of links and a complicated interacting structure that's changing dynamically in time.

156
00:16:02,260 --> 00:16:06,680
So accounting is pretty complicated thing, but at least it's straightforward.

157
00:16:06,680 --> 00:16:12,610
It's clear, it's like physics because there's conservation laws, they're very clear rules about how things work.

158
00:16:12,610 --> 00:16:19,960
Now behaviour. Now we get to the sticky part because behaviour is harder to understand.

159
00:16:19,960 --> 00:16:24,460
I guess I don't know if you can read the caption down there.

160
00:16:24,460 --> 00:16:29,920
The This is a Gary Larson cartoon. Yes, they're all fools.

161
00:16:29,920 --> 00:16:34,690
Yes, they're all fools, gentlemen. But the question remains What kind of fools are they?

162
00:16:34,690 --> 00:16:42,310
No, actually, that's not quite what how economist model behaviour, but I couldn't resist putting it in there the way they model behaviour.

163
00:16:42,310 --> 00:16:46,300
At least the typical model is in terms of rational expectations.

164
00:16:46,300 --> 00:16:51,700
So the concept is there's a rational agent who can compute everything that is a rational agent.

165
00:16:51,700 --> 00:16:57,610
Can you know how well you might? You might or might not have all the information or whatever information they have.

166
00:16:57,610 --> 00:17:03,970
They can do all the computations they need to decide the consequences of that information.

167
00:17:03,970 --> 00:17:14,950
And now, in modern macro and the rational expectations model began in a circa 1960 with John Meuthen had some antecedents before that.

168
00:17:14,950 --> 00:17:20,710
But in modern macro, rational expectation is layered on with frictions.

169
00:17:20,710 --> 00:17:26,830
Now what does that mean? It means that. And by the way, I'm going to I might refer to this as constrained rationality.

170
00:17:26,830 --> 00:17:33,130
So it means that each agent has a utility function that tells them what they like and don't like in a macroeconomic model.

171
00:17:33,130 --> 00:17:40,330
That would mean always that what they want to do is maximise their discounted consumption.

172
00:17:40,330 --> 00:17:47,290
So I want to consume as much as possible so the rational agent will maximise the utility taking others into account.

173
00:17:47,290 --> 00:17:54,160
But the world imposes constraints which are called frictions, which make this harder to do and make that the,

174
00:17:54,160 --> 00:17:58,390
you know, full rationality of this in a sense, less complete.

175
00:17:58,390 --> 00:18:03,850
But it actually makes the problem even harder because it just means there are more constraints they have to take into account.

176
00:18:03,850 --> 00:18:11,740
They can still do all the computation they need to do everything they need to do, and you then call this an equilibrium.

177
00:18:11,740 --> 00:18:15,340
If you're in the state where everybody has maximised their utility,

178
00:18:15,340 --> 00:18:20,590
so everybody finds this place and they go to that state and they stay there, and that would be the equilibrium.

179
00:18:20,590 --> 00:18:28,840
But then stuff happens. So things come in from the outside world economies that the equilibrium zus flows down

180
00:18:28,840 --> 00:18:32,860
a lightning bolt and hits the economy with a shock and knocks it away from equilibrium.

181
00:18:32,860 --> 00:18:39,280
And it's like this rocking horse that sets the horse rocking and it starts to move back towards equilibrium,

182
00:18:39,280 --> 00:18:41,980
but then just throws another shock and knocks it away.

183
00:18:41,980 --> 00:18:48,070
So the world is constantly, never quite at equilibrium because it's constantly getting hit by these shocks.

184
00:18:48,070 --> 00:18:54,280
And now if you look in a production macro model like, say, the nuts and bolts model,

185
00:18:54,280 --> 00:19:03,100
which is one central banks used to think about the economy of you pick a country, they actually had seven different kinds of shocks.

186
00:19:03,100 --> 00:19:07,690
So zoo sector has seven different colours of lightning bolts he can throw out so he can say,

187
00:19:07,690 --> 00:19:15,460
Let's change labour productivity now, let's change the risk perception. Let's change technologies, wages, prices, spending and monetary policy.

188
00:19:15,460 --> 00:19:20,470
Now these are shocks in the sense that no one can know them in advance.

189
00:19:20,470 --> 00:19:28,110
They come from somewhere else outside. It's not the economists job to understand where these come from and this kind of model.

190
00:19:28,110 --> 00:19:32,820
Now, so what's the research programme of modern mainstream macro?

191
00:19:32,820 --> 00:19:41,280
It's basically you take one of the core models, you add a new friction, you solve that model with a new friction, which is hard.

192
00:19:41,280 --> 00:19:47,010
Then you test does that improve the match to empirical facts and then you go back and do it again?

193
00:19:47,010 --> 00:19:54,210
And each one of these steps might take you a year or something. Now what are the problems, in my view and this approach?

194
00:19:54,210 --> 00:20:02,070
First of all, experiments don't support the foundational assumption. You know, they show beyond a shadow of doubt that people aren't very rational.

195
00:20:02,070 --> 00:20:06,720
It's been difficult to match empirical facts. I mean, I've said they are getting better at it.

196
00:20:06,720 --> 00:20:11,820
There's been a lot of work that's gone into this and matched to experimental facts is getting closer.

197
00:20:11,820 --> 00:20:20,250
Still not quite there. And there are debates about whether the frictions that have to be put in or sensible in order to get those matches to happen.

198
00:20:20,250 --> 00:20:28,350
A lot of people think they're not. But in a sense, an even bigger problem is these models are really hard to solve.

199
00:20:28,350 --> 00:20:33,600
And that's not surprising because rationality is a hard thing to do if if you

200
00:20:33,600 --> 00:20:38,170
want to be completely rational and understand everything that can happen to you.

201
00:20:38,170 --> 00:20:45,960
And and what the right thing to do would be in response to those things then and think about everybody else and what they're thinking.

202
00:20:45,960 --> 00:20:49,590
Then you you really have to solve a pretty serious problem.

203
00:20:49,590 --> 00:20:53,460
So what does it mean? It means you have to simplify the assumptions of the model.

204
00:20:53,460 --> 00:21:00,360
You take those two billion balance sheets and you crunch them down into one balance sheet, or maybe a few balance sheets.

205
00:21:00,360 --> 00:21:09,270
But there's no way you can deal with the complexity of the real world in that kind of a setting because it's just too hard to do the calculations.

206
00:21:09,270 --> 00:21:14,070
The other thing that I'll come back to and say more about is rationality suppresses endogenous dynamics.

207
00:21:14,070 --> 00:21:21,180
That is, the reason that the framework is set up this way is because that's what you see emerge in these models.

208
00:21:21,180 --> 00:21:29,010
These models want to sit at rest. Now there are examples of models that have that spontaneously change, but those models are not the norm.

209
00:21:29,010 --> 00:21:33,840
Ninety nine point nine percent of models you find are more are models that sit at risk.

210
00:21:33,840 --> 00:21:43,110
So if you want to explain the dynamics of the economy because the economy is changing all the time, you have to assume that it's coming from shocks.

211
00:21:43,110 --> 00:21:48,540
And finally, it doesn't take advantage of 21st century technology now.

212
00:21:48,540 --> 00:21:57,210
I already said it's hard. I mention that you simplify things down to only looking at a few balance sheets.

213
00:21:57,210 --> 00:22:04,470
So you take all these people on the left and you say, Well, it's too hard to think about all their balance sheets.

214
00:22:04,470 --> 00:22:08,190
So let's do something like make a representative agent like the one on the right.

215
00:22:08,190 --> 00:22:14,150
That's sort of a polyglot mixture of everybody and assume that's the only person in the world or, if you will,

216
00:22:14,150 --> 00:22:19,590
there's seven billion clones of that person and they work for one firm and that

217
00:22:19,590 --> 00:22:24,150
one firm makes one product and that would be the economy and a baseline model.

218
00:22:24,150 --> 00:22:28,200
Now you start to build it back up and the frontier and modern market.

219
00:22:28,200 --> 00:22:35,880
This was, I have to say, the state of the art 30 years ago. The models from 30 years ago have been put into operation of central banks.

220
00:22:35,880 --> 00:22:39,120
Central banks are still using models that are like this.

221
00:22:39,120 --> 00:22:48,480
But if you open, you know, top economics journal and look at the major activity now to try and build models that have heterogeneous agents,

222
00:22:48,480 --> 00:22:53,070
though you also might get a little nervous because those heterogeneous agents sort

223
00:22:53,070 --> 00:22:58,050
of tend to look like more or less educated versions of the guy on the right or,

224
00:22:58,050 --> 00:23:01,590
you know, richer and poorer versions of the guy on the right.

225
00:23:01,590 --> 00:23:07,860
It's a kind of monochromatic notion of diversity, and you might put in a diversity of households,

226
00:23:07,860 --> 00:23:13,200
but you're not going to have a diversity of firms and products and all the other stuff in that same model.

227
00:23:13,200 --> 00:23:18,930
Or you might have a model with some diversity of firms and products, but you're not going to have a diversity of of agents.

228
00:23:18,930 --> 00:23:27,310
So it's still and it's because it's so hard to solve these models. Technology is tough.

229
00:23:27,310 --> 00:23:34,780
Now, this leads to a kind of catch 22, which is I hope I convinced you in the first part that the economy is complex.

230
00:23:34,780 --> 00:23:40,000
It's a complicated two. It's evolving, so it's changing in time all the time.

231
00:23:40,000 --> 00:23:45,370
And that means that data from the distant past isn't very useful from the data in the present.

232
00:23:45,370 --> 00:23:53,020
And that means historical time series or sure, it is also true that it's not very well recorded as you go further back.

233
00:23:53,020 --> 00:24:04,480
So we don't have that much data. In fact, I estimate that the data that goes into calibrating one of these models is about five hundred bytes.

234
00:24:04,480 --> 00:24:16,540
I notice I didn't say megabytes or even kilobytes bytes, not much data that's determining this is the state of the whole world over a 50 year history.

235
00:24:16,540 --> 00:24:20,890
I estimate there's about five hundred bytes of useful information and in those time series.

236
00:24:20,890 --> 00:24:28,390
So there's very little data. So of course you can't. Any statistician knows you can't then estimate a complicated model.

237
00:24:28,390 --> 00:24:35,650
You have too many free parameters. So that means you can only have a simple model, but the economy is complex, so it's a catch 22.

238
00:24:35,650 --> 00:24:43,960
You know, you're damned if you do and damned if you don't. Now, I also mentioned that the standard macros based on old technology,

239
00:24:43,960 --> 00:24:50,620
I've showed a picture there of the computers that existed at the time as basic machinery was getting set up.

240
00:24:50,620 --> 00:24:55,090
So, you know, the data wasn't being collected by the computer is being collected by hand,

241
00:24:55,090 --> 00:24:59,800
and the whole process that set up in national accounting is oriented around

242
00:24:59,800 --> 00:25:05,080
the thinking that was happening like during World War Two and shortly after.

243
00:25:05,080 --> 00:25:12,610
And. And if you look at the way these models are run now, I mean, OK, the computers,

244
00:25:12,610 --> 00:25:17,950
this giant thing in the background, by the way, is vastly less powerful than my iPhone here.

245
00:25:17,950 --> 00:25:25,720
But you know, you can run most macro models on a laptop, and the computing that's actually done is all spin and find this equilibrium.

246
00:25:25,720 --> 00:25:33,730
It's not tracking all the flows of stuff around the globe and the interactions, so it's not staying up to date with technology.

247
00:25:33,730 --> 00:25:40,810
Now what's the alternative? I would argue it's thinking about the economy as a complex system, which has emergent behaviour.

248
00:25:40,810 --> 00:25:44,380
So that's the definition of a complex system, meaning that you have qualitative,

249
00:25:44,380 --> 00:25:48,640
qualitatively different behaviour than that of the individual components.

250
00:25:48,640 --> 00:25:56,230
The classic example would be the human brain, which is made up of neurones actually so many different kinds of neurones,

251
00:25:56,230 --> 00:25:59,020
but each neurone is still a pretty simple device,

252
00:25:59,020 --> 00:26:07,780
and you would never say, Oh, this neurone is conscious or this neurone is, you know, capable of appreciating a poem or something like that.

253
00:26:07,780 --> 00:26:12,400
Somehow, these eighty six billion neurones interact with each other.

254
00:26:12,400 --> 00:26:19,900
And, you know, many trillions of synapses and we end up with thought.

255
00:26:19,900 --> 00:26:25,060
So there's a clear qualitative difference between the lower level component, my level.

256
00:26:25,060 --> 00:26:30,760
You can show that that a necessary condition for that to happen is nonlinear behaviour, the whole it.

257
00:26:30,760 --> 00:26:34,030
That's what allows the whole not to be equal to the sum of the parts.

258
00:26:34,030 --> 00:26:40,180
And you've got to have that if you you're going to get emergence and across lots of fields.

259
00:26:40,180 --> 00:26:47,530
If you want to understand emergence, you have to you have to model on a small scale to see how that emergence comes up for the larger scale.

260
00:26:47,530 --> 00:26:50,860
Now what's the emergence in the economy? That's what I was saying before. You know,

261
00:26:50,860 --> 00:27:03,100
we collectively act in ways that remarkably amplify our ability to do things and and that I would argue the traditional models are not not capturing.

262
00:27:03,100 --> 00:27:10,880
We're not even equipped to talk about in a sensible way. After all, if you're rational when you have to bother.

263
00:27:10,880 --> 00:27:20,930
Evolving and everything else now completes the economics, which is a field I'm working on now, applies complex systems to and methods to economics.

264
00:27:20,930 --> 00:27:27,230
The conceptual model we would argue for is bounded rationality that is saying that reasoning capabilities are limited

265
00:27:27,230 --> 00:27:31,970
and that's very different than constrain rationality because we're not saying that we're perfectly rational.

266
00:27:31,970 --> 00:27:36,950
We're just constrained by stuff. We're saying, actually, there's a limit to how much we can.

267
00:27:36,950 --> 00:27:45,860
We can solve problems and when we're really limited by that and we have to really think collectively about how we collectively solve problems.

268
00:27:45,860 --> 00:27:52,100
Especial is taking that into account, and I'm showing a picture of Herb Simon, who was one of the first people to articulate that idea.

269
00:27:52,100 --> 00:28:00,210
Now I played a bit of a trick on you and my earlier slide when I said I said rationality and I showed you a picture of Bobby Fischer.

270
00:28:00,210 --> 00:28:04,940
Now, in addition, the fact that if you know anything about chess, you know, Bobby Fischer is a nutcase.

271
00:28:04,940 --> 00:28:09,950
But but chess players are, by definition, guardedly rational.

272
00:28:09,950 --> 00:28:14,030
Why? Because nobody can foresee everything in chess, not deep blue.

273
00:28:14,030 --> 00:28:21,530
Not Bobby Fischer. Not anybody. Because chess is too hard. I think there's a lesson there now.

274
00:28:21,530 --> 00:28:29,090
But on the other hand, there are situations where, say, older children say playing the game of knots and crosses can become rational.

275
00:28:29,090 --> 00:28:34,340
Like, I remember when I was about 10 OK, in America, we call it Tic TAC Toe.

276
00:28:34,340 --> 00:28:38,840
We were playing Tic TAC Toe. It seemed like a lot of fun. You know, when sometimes it was really fun.

277
00:28:38,840 --> 00:28:44,300
And then I figured out that there was a strategy that I could always at least get a draw.

278
00:28:44,300 --> 00:28:50,870
And then my friends figured it out, maybe a day later and it became hopelessly boring and we quit.

279
00:28:50,870 --> 00:28:54,500
Now what happened? We found an equilibrium. We became rational.

280
00:28:54,500 --> 00:28:59,510
We learnt enough because the game simple to actually be rational with respect to this game.

281
00:28:59,510 --> 00:29:03,680
So there are situations where rationality makes sense to explain things.

282
00:29:03,680 --> 00:29:12,080
But the problem gets harder. It doesn't happen. Now I'm going to give an empirical test, so I'm going to ask you all to play a beauty contest game.

283
00:29:12,080 --> 00:29:16,700
So the idea is to get a number between zero and 100.

284
00:29:16,700 --> 00:29:22,020
That's two thirds of the average guess. And I expect there's a few wise guys that understand that.

285
00:29:22,020 --> 00:29:28,040
So keep your mouth shut. But I want you to all think about this because I want everybody to get the number.

286
00:29:28,040 --> 00:29:33,110
So thing I'll remember. Everybody's going to pick a number between zero and 100,

287
00:29:33,110 --> 00:29:38,000
and the winner will be the one that gets two thirds of the average guess of everybody else in the room.

288
00:29:38,000 --> 00:29:44,870
Now the little bit, while you think about it for a minute, little picture on the side is say this is inspired by Canes,

289
00:29:44,870 --> 00:29:52,330
who pointed out that the stock market is like a beauty contest because in that day there were beauty contests where you would enter a lottery.

290
00:29:52,330 --> 00:29:55,430
You would have to pick one of these women as the most beautiful,

291
00:29:55,430 --> 00:29:59,810
which actually means the one you think other people will think is going to be the most beautiful.

292
00:29:59,810 --> 00:30:03,290
And that's what's important if you want to win the contest. So it's not beauty.

293
00:30:03,290 --> 00:30:07,790
That's there's no objective standard for that. It's a question of understanding what people perceive as beautiful.

294
00:30:07,790 --> 00:30:11,810
So it's a subjective game. Now everybody have their guests. All right.

295
00:30:11,810 --> 00:30:17,030
Now I'm going to show the next slide so you can see what a large group of people tend to do on this.

296
00:30:17,030 --> 00:30:27,440
And this was a survey done by a Danish newspaper. And so you see on the x axis the frequency of the various guesses of 19000 respondents

297
00:30:27,440 --> 00:30:31,700
to this daily Danish newspapers query had more time to think about it than you did.

298
00:30:31,700 --> 00:30:34,880
And so you can see the guesses that everybody made.

299
00:30:34,880 --> 00:30:45,470
Now, when you look at this, you'll notice some disturbing things like if you go over here, there's a bunch of people that just more than two thirds.

300
00:30:45,470 --> 00:30:50,600
But just think about this for a minute. Let's suppose everybody in the room had just one hundred.

301
00:30:50,600 --> 00:30:55,070
What would the and what would the winner be? It would be sixty six.

302
00:30:55,070 --> 00:30:58,370
OK, so the answer's got to be less than that. All right.

303
00:30:58,370 --> 00:31:03,110
No. If everybody guessed it random, so everybody just picked a number out of the hat.

304
00:31:03,110 --> 00:31:07,550
Then the average would be 50. And so the winner would be a third.

305
00:31:07,550 --> 00:31:11,540
And actually, you see the most people voted for a third, so people aren't.

306
00:31:11,540 --> 00:31:16,340
The reasoning here? Right. More than six percent of the people voted for a third.

307
00:31:16,340 --> 00:31:21,080
But then there were those people who said, Let's go one more level because I'll bet that,

308
00:31:21,080 --> 00:31:25,820
you know, let's go to level two because if they're guessing a third.

309
00:31:25,820 --> 00:31:32,270
Then let's take two thirds of a third. So they just said, What is that two nights?

310
00:31:32,270 --> 00:31:36,230
And so they guessed two nights or so. And in fact, that's the winner.

311
00:31:36,230 --> 00:31:42,830
But then it goes down and there's the people here who just 0.2 percent of the people just rule, which if you think about it,

312
00:31:42,830 --> 00:31:50,660
if you'd kept carrying that process through, you would have ended up at zero, which is the Nash equilibrium for this game.

313
00:31:50,660 --> 00:31:54,500
And because this set of decisions that everybody had gone to zero,

314
00:31:54,500 --> 00:32:01,310
then if everybody's a winner at that point and if anybody deviates, they're going to, they're not going to be a winner anymore.

315
00:32:01,310 --> 00:32:07,400
So that's the whole idea of a national government. Equilibrium economics tend to have this flavour now.

316
00:32:07,400 --> 00:32:21,480
OK, so how useful was that idea? So the game theorists and economists just zero with a losing get right is a bad model of human behaviour.

317
00:32:21,480 --> 00:32:25,320
Now, of course, the sceptic could say What if we kept playing this game?

318
00:32:25,320 --> 00:32:33,540
Let's play it again and again. Well, in fact, it would start moving towards zero because you can see here already people are, you know,

319
00:32:33,540 --> 00:32:39,870
there's enough intelligence in the room that there are a lot more maths on this side of the histogram than on that one.

320
00:32:39,870 --> 00:32:47,130
And so if we played this game again and again, your guesses would start going down and we might actually well converge on zero.

321
00:32:47,130 --> 00:32:48,840
But what about other games?

322
00:32:48,840 --> 00:32:58,980
So I devoted the last ten or not all of it, but off and on for 10 or 15 years with some really great collaborators who are listed there.

323
00:32:58,980 --> 00:33:03,340
And we've written about four or five papers studying this and lots of other kinds of games.

324
00:33:03,340 --> 00:33:09,450
We've actually now exhaustively studied normal form games, and we've shown which is a particular kind of game.

325
00:33:09,450 --> 00:33:15,990
You don't need to worry about it here. But we've shown that equilibrium becomes unlikely when the games are complicated and competitive.

326
00:33:15,990 --> 00:33:20,160
That is, as soon as you have more than two players and more than a few possible actions.

327
00:33:20,160 --> 00:33:25,680
And when the incentives of the players aren't lined up so that if I win, you tend to lose and vice versa,

328
00:33:25,680 --> 00:33:31,320
then you don't go to equilibrium anymore under, you know, well, substantiated learning algorithms.

329
00:33:31,320 --> 00:33:36,540
And instead, I'm sorry, everybody can't see danger of two screens.

330
00:33:36,540 --> 00:33:44,370
But as the game gets more complicated when the game's competitive, we quickly go up to a place where we will almost never go to equilibrium.

331
00:33:44,370 --> 00:33:48,840
And instead, what do we see? We see chaotic dynamics happening.

332
00:33:48,840 --> 00:33:55,890
So right now complex the economics, we try to take behavioural economics seriously.

333
00:33:55,890 --> 00:33:59,730
Now I'm embarrassed to say that Sanjay Dummy was giving us the talk today,

334
00:33:59,730 --> 00:34:04,320
and he kind of said, Well, look, you guys aren't really taking it seriously. Well, we at least intend to.

335
00:34:04,320 --> 00:34:06,600
And me too. And I think I'll come back in a minute to say why.

336
00:34:06,600 --> 00:34:17,040
I think that maybe isn't as good as it could be, you know, but we do believe in models where the agents reason and in a empirically justifiable way,

337
00:34:17,040 --> 00:34:21,870
they follow heuristics, they do some myopic reasoning, just like the people who won that game.

338
00:34:21,870 --> 00:34:30,030
Did, they update the heuristics that recently are working well and they may or may not converge to equilibrium?

339
00:34:30,030 --> 00:34:33,300
Now, in order to do this, we have to simulate the world,

340
00:34:33,300 --> 00:34:40,110
meaning we actually use a computer to simulate what people are doing because it's too complicated to do.

341
00:34:40,110 --> 00:34:46,710
Using standard, you know, close to maths like you get taught in in a maths course in college.

342
00:34:46,710 --> 00:34:52,680
And so we mimic the world on a computer and.

343
00:34:52,680 --> 00:34:58,860
Bounded rationality is actually essential for us because it means we can still simulate things in complicated situations,

344
00:34:58,860 --> 00:35:04,020
just like real people can make decisions in complicated situations.

345
00:35:04,020 --> 00:35:12,330
It's not true for constrained rationality. There's also a difference in philosophy because in complexity economics, we we reject as if reasoning.

346
00:35:12,330 --> 00:35:16,750
I mean, even if rationality with frictions isn't literally what people do,

347
00:35:16,750 --> 00:35:24,600
maybe it's the economy behaves as if they did to say, Well, actually, let's just use the principle of Earth similitude.

348
00:35:24,600 --> 00:35:30,480
Let's try and make things kind of like the real world and at least the essential features and are.

349
00:35:30,480 --> 00:35:35,430
And so we should have assumptions that are plausible and that we can empirically verify following,

350
00:35:35,430 --> 00:35:40,480
of course, Einstein's dictum that everything should be made as simple as possible, but no simpler.

351
00:35:40,480 --> 00:35:46,340
And now I don't have a lot of time left, and I want to make sure I leave plenty of time for discussion, but I'm going to give you.

352
00:35:46,340 --> 00:35:53,830
So I first want to give an example of how you immediately get endogenous dynamics as soon as you deviate from the standard model.

353
00:35:53,830 --> 00:35:59,440
But we're going to start with a standard model. And these are my collaborators at the top led by Ukiah Sano.

354
00:35:59,440 --> 00:36:05,680
I think I over there. And and so we take a standard model with a representative household.

355
00:36:05,680 --> 00:36:10,240
This guy in the picture, we got a household that has to make one decision.

356
00:36:10,240 --> 00:36:15,820
That is, how much should I save and what doesn't get saved,

357
00:36:15,820 --> 00:36:23,200
what gets saved then gets invested in producing stuff for the economy and the rest gets consumed.

358
00:36:23,200 --> 00:36:28,100
So you can either eat your earnings or you can reinvest them to make more stuff.

359
00:36:28,100 --> 00:36:32,830
So you have something the next period and you've got to find the right compromise.

360
00:36:32,830 --> 00:36:38,530
Now in the standard model, and you know, you may not follow this because particularly since I can't point at both things at once,

361
00:36:38,530 --> 00:36:43,720
but there is a there's a this is showing the flow of how things flow around your.

362
00:36:43,720 --> 00:36:48,940
Let's suppose you're making a certain you're using a certain savings rate.

363
00:36:48,940 --> 00:36:56,530
How much capital do you do you have accumulate versus how much are you consuming at each time step?

364
00:36:56,530 --> 00:37:04,210
And how do you move around the space as you make these rational decisions that this agent is doing?

365
00:37:04,210 --> 00:37:14,120
And what you see is that first of all, everything flies off to infinity and there's only one point here, and that's a fix.

366
00:37:14,120 --> 00:37:15,190
Well, there's a fixed point here.

367
00:37:15,190 --> 00:37:21,580
There's a fixed point down here where the whole economy collapses and there's a fixed point here where things don't move, but it's unstable.

368
00:37:21,580 --> 00:37:24,730
Any little perturbation away from that point will set it off.

369
00:37:24,730 --> 00:37:34,240
So how does this work in the standard economics model you assume this household is doing here is so smart that they managed to sit at that place.

370
00:37:34,240 --> 00:37:37,900
And then you can OK, the rest of the model proceed to get hit by a shock.

371
00:37:37,900 --> 00:37:42,720
They move back to the place, gets hit by a shock and move back now.

372
00:37:42,720 --> 00:37:49,950
We changed it to say, well, let's suppose we have a bunch of households and they have some kind of social network.

373
00:37:49,950 --> 00:37:55,860
And what happens is once in a while a household wakes up, looks around at his neighbours or her neighbours and says,

374
00:37:55,860 --> 00:38:01,320
Oh, that neighbour is the one who's consuming the most because everybody's been conspicuous about their consumption.

375
00:38:01,320 --> 00:38:05,700
And and I'll just consume whatever they're consuming. I'll actually know.

376
00:38:05,700 --> 00:38:09,810
I will choose their savings rate because it worked for them. So we do that.

377
00:38:09,810 --> 00:38:12,990
And what happens then? Well, take this picture here.

378
00:38:12,990 --> 00:38:19,860
Instead of sitting at the fixed point, OK, in some cases, we see it sitting at a fixed point that's highly inefficient.

379
00:38:19,860 --> 00:38:26,280
As we lengthen the period out, it gets more efficient, meaning more stuff gets produced and households better off.

380
00:38:26,280 --> 00:38:33,480
But then finally, what happens when it starts oscillating? So here you see an orbit where the the colour is, the savings rate.

381
00:38:33,480 --> 00:38:36,390
You see, we're orbiting around that fixed point now.

382
00:38:36,390 --> 00:38:43,300
So you're getting spontaneous dynamics that's coming out of nowhere because all these agents are doing is waking up and looking at each other.

383
00:38:43,300 --> 00:38:50,700
There's no shocks in the normal sense here, other than a little bit of randomness and when the agents wake up to look at the world.

384
00:38:50,700 --> 00:38:53,970
It just comes out of the endogenous dynamics of this model.

385
00:38:53,970 --> 00:39:00,450
And you see that we get something that looks an awful lot like a business cycle because we see, you know,

386
00:39:00,450 --> 00:39:07,530
the output of the economy now spontaneously fluctuates in time as the average savings rate of all these agents fluctuates.

387
00:39:07,530 --> 00:39:13,050
And as each of them makes their decisions, each agent actually is having their savings rate fluctuate.

388
00:39:13,050 --> 00:39:18,750
Now, one way to think about what's going on here is there's something, I think, a better model than the rocking horse.

389
00:39:18,750 --> 00:39:23,370
It's actually a pull balancing. In the old days, I would have had a pointer that I could use.

390
00:39:23,370 --> 00:39:29,070
But if you imagine trying to balance a pole in your hand, it turns out of the poles shorter than about three feet.

391
00:39:29,070 --> 00:39:32,460
You can't do it. As soon as the pole gets long enough, you can do it.

392
00:39:32,460 --> 00:39:37,620
You can just hold it more or less upright, but it won't sit exactly upright.

393
00:39:37,620 --> 00:39:42,060
And now how does a traditional macro model model this?

394
00:39:42,060 --> 00:39:50,370
This guy assumes it stays exactly vertical unless it gets hit by a shark, which would knock the pole balance or his hand.

395
00:39:50,370 --> 00:39:58,410
The whole the pole balance would make the optimal movement to set the pole back up to being vertical again until the next shock you.

396
00:39:58,410 --> 00:40:03,840
But of course, that's not the way this works. The way it works is that you make a correction.

397
00:40:03,840 --> 00:40:07,440
The corrections imperfect pole overreacts, you overreact.

398
00:40:07,440 --> 00:40:14,620
So there's constantly indigenous populations that are chaotic in nature, just as we saw in our modified model.

399
00:40:14,620 --> 00:40:19,140
Now I'm got to run out of time, so I'm going to go quickly here.

400
00:40:19,140 --> 00:40:26,070
Now I'm just going to give some examples to demonstrate that economics can be done without equilibrium going to fly through these at blinding speed.

401
00:40:26,070 --> 00:40:30,210
The purpose is just to show it so I can make my my final remarks and it will go on.

402
00:40:30,210 --> 00:40:36,630
So these are all done with no utility functions, no rational agents and no perfect maximiser.

403
00:40:36,630 --> 00:40:42,360
So we made a model of the crisis of 2008, where we just model the behaviour that we know that people were doing.

404
00:40:42,360 --> 00:40:43,800
They were following value at risk.

405
00:40:43,800 --> 00:40:51,330
They were using historical average of volatility to compute the risk and adjusting their leverage in the market accordingly.

406
00:40:51,330 --> 00:40:57,390
And we see a spontaneous oscillation come out of our model that has a period of about 10 years,

407
00:40:57,390 --> 00:41:01,890
with the prices in the stock market going up and down and leverage going up and down.

408
00:41:01,890 --> 00:41:05,100
We have models of this financial stability of the European banking system,

409
00:41:05,100 --> 00:41:12,360
where we're literally taking the balance sheets of the 100 most systemically important banks in

410
00:41:12,360 --> 00:41:19,410
the UK and looking at the way they interact when they do things to try and do their risk control,

411
00:41:19,410 --> 00:41:25,900
which can then set up large scale oscillations due to systemic effects.

412
00:41:25,900 --> 00:41:31,260
A team of us are looking at unemployment, making agent based model of job transitions,

413
00:41:31,260 --> 00:41:36,510
where we can predict the effect of an automation shock and and think about questions like which professions are

414
00:41:36,510 --> 00:41:42,420
safest because it might be even though your profession is not likely to get automated or is likely to get automated.

415
00:41:42,420 --> 00:41:46,560
Some professions may have an easy time transitioning other ones that are and vice versa.

416
00:41:46,560 --> 00:41:52,590
And if you come in two weeks, Penny may well at least say something about automation in the job space.

417
00:41:52,590 --> 00:41:57,300
Thinking about technological change and taking advantage of these facts, I have you.

418
00:41:57,300 --> 00:42:02,310
I showed you things that are highly heterogeneous, but I don't know if you noticed this in this picture,

419
00:42:02,310 --> 00:42:07,980
but there are amazingly persistent once you go down to the level of individual technologies.

420
00:42:07,980 --> 00:42:13,740
It's not like hospital services got really more expensive one year and then less expensive the next year.

421
00:42:13,740 --> 00:42:22,620
For 20 years, they got more expensive every year. Televisions got cheaper every year, and these patterns were amazingly persistent.

422
00:42:22,620 --> 00:42:30,510
But to see that persistence, you have to get down to the micro level, and normal models are not doing that.

423
00:42:30,510 --> 00:42:34,140
This allows us to do things like forecast technological progress,

424
00:42:34,140 --> 00:42:39,390
and we can even forecast how good our forecast is and we can put all these forecasts

425
00:42:39,390 --> 00:42:43,380
together and think about how the different components in that picture I showed you are

426
00:42:43,380 --> 00:42:48,690
interacting with each other and how technological progress in one domain is is percolating

427
00:42:48,690 --> 00:42:53,100
over to technological progress in other domains and use this to predict growth.

428
00:42:53,100 --> 00:43:02,970
We can bring in ideas from ecology to organise firms and industries into ecological structures that that use a

429
00:43:02,970 --> 00:43:10,660
totally literal analogy to trophic levels in ecology that actually turns out to be very closely related to growth.

430
00:43:10,660 --> 00:43:16,800
OK, we can make predictions without and and we can use this to compute the cost of the green energy transition.

431
00:43:16,800 --> 00:43:25,860
And one of the results that we hope to release a paper with in the next month is that actually the green energy transition is not just cheap,

432
00:43:25,860 --> 00:43:31,230
it's probably actually when net present value is taken into account at just about any discount

433
00:43:31,230 --> 00:43:35,940
rate likely to be cheaper than the alternative of just sticking with business as usual.

434
00:43:35,940 --> 00:43:36,480
And again,

435
00:43:36,480 --> 00:43:43,990
we can do that because we get down to the fine grain level where we can really think about the individual technologies and put them together.

436
00:43:43,990 --> 00:43:53,200
Now, I also think we can solve a Catch 22 macro by going to what I'm calling global microeconomics that is letting macro

437
00:43:53,200 --> 00:44:01,390
emerge from micro taking advantage of heterogeneity rather than viewing it as something that makes things hard.

438
00:44:01,390 --> 00:44:06,370
Taking advantage of the fact that there's a lot more data micro scales that allows

439
00:44:06,370 --> 00:44:11,710
us to get better statistical significance and deal with the Catch 22 of macro,

440
00:44:11,710 --> 00:44:15,790
I think we will naturally see the emergence of more endogenous dynamics.

441
00:44:15,790 --> 00:44:19,720
We can then model that emergence and because we're modelling in a finer scale,

442
00:44:19,720 --> 00:44:23,830
we have more things to predict, which means we have more ways to validate our model.

443
00:44:23,830 --> 00:44:27,730
Now, just to wrap up, I mean, complexity economics is yawn.

444
00:44:27,730 --> 00:44:31,810
So I'm in that last slide. I'm throwing out a super ambitious programme.

445
00:44:31,810 --> 00:44:37,870
But you know, I estimate there have been the order of five hundred person years of work that's gone into this field.

446
00:44:37,870 --> 00:44:45,880
In contrast to, say, the 50000 that have gone into the mainstream approach, a lot of work remains to be done.

447
00:44:45,880 --> 00:44:50,020
We need to develop new methods. We need to fit models to a time series better.

448
00:44:50,020 --> 00:44:55,030
We need to do better parallelism so we can run big models.

449
00:44:55,030 --> 00:44:58,780
We need to create standard software libraries to gather better data sets.

450
00:44:58,780 --> 00:45:06,130
We need to pay more careful attention to behavioural economists. There's a huge amount to be done, but I think there's a lot of promise now.

451
00:45:06,130 --> 00:45:09,670
The mainstream is resisting this. I think for several reasons.

452
00:45:09,670 --> 00:45:16,810
One is it requires abandoning foundational assumptions that have been used in use since the middle of the 20th century.

453
00:45:16,810 --> 00:45:21,730
So it's a pretty big shift in what has to happen, requires a very different toolkit,

454
00:45:21,730 --> 00:45:25,870
a different set of skills, and it requires a different attitude about science.

455
00:45:25,870 --> 00:45:28,180
It's a different scientific method.

456
00:45:28,180 --> 00:45:36,160
And anybody who wants to learn more, I give the URL for my website where you can see the draught of the introduction of my book.

457
00:45:36,160 --> 00:45:41,050
It's still very hot and in flux.

458
00:45:41,050 --> 00:45:42,670
But if you want to take a look, you can look there.

459
00:45:42,670 --> 00:45:49,990
And if you want to see any details about the papers I flew through at blinding speed, you can look on our website there.

460
00:45:49,990 --> 00:46:08,330
Thank you very much. So that's a real tour de force, other questions for day.

461
00:46:08,330 --> 00:46:12,890
We'd like to go first, Tim, if you could wait for the microphone.

462
00:46:12,890 --> 00:46:19,610
I also remind you that this is being webcast, so when you ask a question, just be aware of that.

463
00:46:19,610 --> 00:46:32,720
Don, I wanted to ask. I just actually picking up on your question whether there is any similarity between your and my field in the following sense.

464
00:46:32,720 --> 00:46:36,830
You know, so I'm a climate physicist, climate modeller.

465
00:46:36,830 --> 00:46:44,600
If you want to know if you know, to model the warming of the planet as a whole due to increase in carbon dioxide,

466
00:46:44,600 --> 00:46:50,870
you can use very simple analytic models, which essentially are looking for new equilibrium.

467
00:46:50,870 --> 00:46:54,770
It's a kind of and it's a reasonably elegant and simple approach.

468
00:46:54,770 --> 00:46:58,010
But if you want to know what's happening regionally, in my view,

469
00:46:58,010 --> 00:47:06,140
there is no substitute than actually modelling the whole climate system with all its complications and interactions and so on.

470
00:47:06,140 --> 00:47:13,280
But there is a there is a subset of my field that views that to be somehow inelegant brute force.

471
00:47:13,280 --> 00:47:19,820
It doesn't have that mathematical panache, that low water, you know,

472
00:47:19,820 --> 00:47:30,110
semi analytic models have an I try to argue to these people that you that that view is very misguided and there is a role for elegant mathematics,

473
00:47:30,110 --> 00:47:34,970
but it's not him. So I'm just wondering, does that play that just sort of sociology?

474
00:47:34,970 --> 00:47:39,050
Does that play a role in your field as well as places to rest like brute force?

475
00:47:39,050 --> 00:47:46,940
It plays a huge role. I mean, I think actually, if I, I should probably added that on here that there's if you if you grow up doing

476
00:47:46,940 --> 00:47:51,560
mathematics and the whole thing you learn how to do is to solve equations,

477
00:47:51,560 --> 00:47:56,750
then somebody says, well, actually forget about that and go to stimulate stuff on a computer.

478
00:47:56,750 --> 00:48:01,730
OK. Fermi might have been, you know, Fermi was one of the first people to reprogramme the first computers,

479
00:48:01,730 --> 00:48:06,290
but but most it's a whole different kind of ballgame.

480
00:48:06,290 --> 00:48:12,170
I think there is an important question of how far can you get within our genius type, you know, aggregate model.

481
00:48:12,170 --> 00:48:18,620
And I don't want to say that you can't do anything with an aggregate macro model, but if there are nonlinear interactions, then you may miss that.

482
00:48:18,620 --> 00:48:26,060
So for example, if it turns out, if it had turned out that we're in serious danger of reversing the Great Atlantic salt current,

483
00:48:26,060 --> 00:48:29,270
then our genius prediction might have been really, really wrong.

484
00:48:29,270 --> 00:48:36,140
Because when you start looking at the structure, you have this big nonlinear effect that sends the system into a whole different state.

485
00:48:36,140 --> 00:48:37,940
And so I think we have to worry about that.

486
00:48:37,940 --> 00:48:44,600
And I think actually in the economy, we're seeing oscillations happening, the system that are going on all the time.

487
00:48:44,600 --> 00:48:49,580
So it's a little bit like, you know, studying business cycles as opposed to saying climate.

488
00:48:49,580 --> 00:48:53,610
I mean, business cycles might be more like weather as opposed to climate.

489
00:48:53,610 --> 00:48:58,340
And so for weather, you really can't use an aggregate model, right?

490
00:48:58,340 --> 00:49:05,450
Weather is all about simulating and fine scale. And as you know, it's limited by how good your computer is and how much data you have.

491
00:49:05,450 --> 00:49:11,930
So but it's a very, very important point. Thank you.

492
00:49:11,930 --> 00:49:17,180
How do you judge? I'm sorry if you could just wait for the microphone.

493
00:49:17,180 --> 00:49:21,290
How do you judge whether or not your models are successful? Yeah.

494
00:49:21,290 --> 00:49:25,820
So I'm a firm believer in empirical judgement of success.

495
00:49:25,820 --> 00:49:32,000
Now that said, you, you often need an incubation period to really get success to happen.

496
00:49:32,000 --> 00:49:38,000
And and there's a question of how much success can you get with small models doing this kind of thing?

497
00:49:38,000 --> 00:49:44,150
I think you can get some and I think I mean, I could have, you know, had a bunch of slides of the things I would rate as successes.

498
00:49:44,150 --> 00:49:49,790
And if you read my book, I spent a lot of time talking about that. But I think there's still relatively small successes.

499
00:49:49,790 --> 00:49:58,580
We still have no model that can step in and replace the DSG model for, say, modelling the British economy or the world economy.

500
00:49:58,580 --> 00:50:05,540
But so ultimately, it needs to be empirical and it should be asking, what can we predict and how what we do we predicted.

501
00:50:05,540 --> 00:50:09,470
But to get there, we may need an incubation period where we say, Well,

502
00:50:09,470 --> 00:50:16,700
this looks promising and that looks promising to come back to whether, you know, the first physical weather forecast on a computer.

503
00:50:16,700 --> 00:50:24,020
We're done in 1950. But it was 30 years before they can beat subjective weather forecasters who just spoke their periods and said,

504
00:50:24,020 --> 00:50:29,300
This is what I think the weather will do tomorrow. Now, once they beat them, they rocketed past them.

505
00:50:29,300 --> 00:50:34,910
And weather forecasting is now far, far better than it was in 1980.

506
00:50:34,910 --> 00:50:40,940
But it took 30 years of hard work and billions of dollars of investment to get the question there.

507
00:50:40,940 --> 00:50:45,420
And then they. Thank you. I appreciate the question.

508
00:50:45,420 --> 00:50:51,120
Looking at ecological and economic modelling together, what about political modelling?

509
00:50:51,120 --> 00:50:56,910
For example, you might say a macroeconomic trend is income inequity.

510
00:50:56,910 --> 00:51:03,150
And so these representative people that you're creating are going to be more bipolar.

511
00:51:03,150 --> 00:51:07,220
So can you can you model an income inequity into into this complex?

512
00:51:07,220 --> 00:51:13,890
You know what, one of the places that I, to be honest, I don't think this approach is not realised.

513
00:51:13,890 --> 00:51:19,200
Its potential is in thinking about inequality. So it's very easy to just go back to the, you know,

514
00:51:19,200 --> 00:51:24,840
and have as many different groups of people as you want to have match it all up against empirical data.

515
00:51:24,840 --> 00:51:30,570
And now you have to have a model of how those people are going to behave as they as time marches along.

516
00:51:30,570 --> 00:51:38,340
But you can simulate, you know, the economy and the feedbacks and and what different what effect different policies will have on inequality.

517
00:51:38,340 --> 00:51:44,070
And I think this is one of the major unexploited strengths of this approach.

518
00:51:44,070 --> 00:51:50,760
But unexploited because we really haven't done that very well. They're very simple tinker toy models that do show immediately.

519
00:51:50,760 --> 00:51:55,590
I mean, you automatically in these models get inequality. You distort some agents off.

520
00:51:55,590 --> 00:52:00,750
You let them go, boom. You get it. You get a distribution actually even looks more or less right.

521
00:52:00,750 --> 00:52:05,850
You get a log normal out more or less right away. You can even get the tail on the log normal.

522
00:52:05,850 --> 00:52:12,510
But but then if you really want to set it up so you can take a policy and say, what will the policy do to that?

523
00:52:12,510 --> 00:52:22,710
And how will this distribution chain feed back into the global GDP and so forth?

524
00:52:22,710 --> 00:52:26,550
That's a much harder problem and hasn't been done. OK, thank you.

525
00:52:26,550 --> 00:52:34,140
I just wondered what what they might be able said something about the role of technology, because it seems to me that in your graph,

526
00:52:34,140 --> 00:52:45,330
where you have areas of consistent increase in prices or cost, that in actual fact, the more the less you're able to code and automate knowledge.

527
00:52:45,330 --> 00:52:51,510
So the more tacit the knowledge and the more interactive the area, the more you're likely to see increasing prices.

528
00:52:51,510 --> 00:52:55,690
What are your thoughts on that? Well, I don't know.

529
00:52:55,690 --> 00:53:03,450
OK. There is there is a long known fact that, you know, was called balls disease disease kind of named almost at the top of,

530
00:53:03,450 --> 00:53:12,990
you know, what's the long term improvement rate of services versus, you know, manufacturing?

531
00:53:12,990 --> 00:53:23,250
And and it is true that services generally improve at a slower rate than the more or high tech ends of manufacturing like computers.

532
00:53:23,250 --> 00:53:29,490
So that seems to be an empirical regularity. It's interesting to try and understand why.

533
00:53:29,490 --> 00:53:33,930
I don't think there are really good explanations for that, although, you know,

534
00:53:33,930 --> 00:53:39,840
people kind of understand why Moore's law would have a 40 percent per year improvement rate for so long.

535
00:53:39,840 --> 00:53:44,910
But but in general, it's this is a big unknown about how do these different technologies?

536
00:53:44,910 --> 00:53:50,100
Why are televisions improving so much faster than hospital services?

537
00:53:50,100 --> 00:53:54,090
And you can tell individual stories about some of these.

538
00:53:54,090 --> 00:53:59,730
But but actually having a broad theory that tells us about this is one of the objectives of what we're trying to do in our group.

539
00:53:59,730 --> 00:54:04,380
And I'd be lying if I said we understood at this point.

540
00:54:04,380 --> 00:54:12,570
Thank you. I'd say you alluded to the idea that complexity economics recognises that the economy is more networks

541
00:54:12,570 --> 00:54:20,100
than conventional economic approach is a bit like a local social network around investor decisions,

542
00:54:20,100 --> 00:54:29,640
for example. So I was wondering if you could talk a little bit about the extent to which I understand you network dynamics is up to

543
00:54:29,640 --> 00:54:37,110
scratch or what you would like to see in terms of development and network understandings that advance complex economics.

544
00:54:37,110 --> 00:54:47,430
So can I sort of lived through the revolution in networks? You know, when I was a graduate student, there was a field called graph theory.

545
00:54:47,430 --> 00:54:52,230
And, you know, I spent some time staring at books and graph theory, and it looked really cool.

546
00:54:52,230 --> 00:54:56,580
And then people started realising less well. So the sociologists first, actually.

547
00:54:56,580 --> 00:55:04,050
But then physicists jumped in. I'm good friends with several of them and began saying, Look, let's apply this to the real world.

548
00:55:04,050 --> 00:55:13,140
We really need to take this stuff out of the mathematical textbooks and put it to data and develop techniques for things like understanding,

549
00:55:13,140 --> 00:55:18,450
clustering and, you know, getting a clearer picture of descriptive picture of networks.

550
00:55:18,450 --> 00:55:22,740
Then the sort of next step is to start thinking about dynamics on the networks,

551
00:55:22,740 --> 00:55:26,880
like the example I showed you with yuki's model with, you know, the there are there.

552
00:55:26,880 --> 00:55:32,670
We have thousands of households that are forming a network and there are some dynamics on that network.

553
00:55:32,670 --> 00:55:35,790
Very simple because they just copy each other and adjust your savings rate.

554
00:55:35,790 --> 00:55:45,510
But there's a dynamics and agent based modelling as it gets more serious, is essentially just getting more realistic about those processes.

555
00:55:45,510 --> 00:55:52,170
Causation These models typically live on networks because you have a bunch of agents who are the nodes and the links or their interactions,

556
00:55:52,170 --> 00:55:57,120
then they have several different kinds of links and they may move around in time and so on.

557
00:55:57,120 --> 00:56:00,540
But that's really what's going on in an agent based model. And really,

558
00:56:00,540 --> 00:56:10,560
what we're talking about as we develop the whole field is going from just the descriptive network stuff that was developed 15 years ago or so,

559
00:56:10,560 --> 00:56:15,930
20 years ago now to actually dealing with dynamic models that live on these networks.

560
00:56:15,930 --> 00:56:20,970
Now I actually have to mention that I built a model back and started in 86,

561
00:56:20,970 --> 00:56:26,070
and it got published in about 1990, where we literally did have a dynamic network.

562
00:56:26,070 --> 00:56:31,170
We had dynamics on the network. We call this mother dynamics. It was the origin of life.

563
00:56:31,170 --> 00:56:34,800
And so I did put a pole in the ground deck there a long time ago.

564
00:56:34,800 --> 00:56:42,150
The field has come a long way since then. Sorry, but we're going to have to bring things to an end with that, so I apologise for that.

565
00:56:42,150 --> 00:56:49,380
This is the first and several in five talks we're having in this series called Evolving Economics or our next speaker,

566
00:56:49,380 --> 00:56:54,480
which you've heard about several times, is Penny Mealy, and that will be in two weeks time.

567
00:56:54,480 --> 00:56:58,920
And penny, we'll be talking about navigating knowledge, new new tools for the economy.

568
00:56:58,920 --> 00:57:24,564
So please, could you join me for something we're thinking going again for a really?