1
00:00:02,720 --> 00:00:16,360
Speaker 1: Bloomberg Audio Studios, Podcasts, Radio News.

2
00:00:18,520 --> 00:00:22,480
Speaker 2: Hello and welcome to another episode of The Odd Lots Podcast.

3
00:00:22,560 --> 00:00:24,960
Speaker 3: I'm Joe Wisenthal and I'm Tracy Alloway.

4
00:00:25,239 --> 00:00:28,080
Speaker 2: Tracy, one of the things that I think we like

5
00:00:28,200 --> 00:00:32,400
Speaker 2: to do on this podcast is sort of deabstract the

6
00:00:32,440 --> 00:00:35,640
Speaker 2: things that we take for granted in the world. There

7
00:00:35,640 --> 00:00:38,600
Speaker 2: are various processes. We always say this when there's like

8
00:00:38,640 --> 00:00:41,160
Speaker 2: a blow up or something like that, where it's like, oh,

9
00:00:41,320 --> 00:00:43,800
Speaker 2: if we're having to pay attention to X or y,

10
00:00:43,880 --> 00:00:46,200
Speaker 2: there must be something going on. But we don't always

11
00:00:46,200 --> 00:00:48,000
Speaker 2: have to wait for blow ups. But like, we live

12
00:00:48,040 --> 00:00:51,560
Speaker 2: in this world where, like you click buttons and things happen,

13
00:00:52,000 --> 00:00:54,560
Speaker 2: and you have some intuition of what happens is after

14
00:00:54,600 --> 00:00:56,520
Speaker 2: the button is clicked, but you don't really have a

15
00:00:56,560 --> 00:00:59,760
Speaker 2: great intuition of like what happens between the button clicking

16
00:00:59,840 --> 00:01:00,840
Speaker 2: and the thing happening.

17
00:01:00,960 --> 00:01:04,040
Speaker 3: Absolutely, I actually don't know how the process of like

18
00:01:04,360 --> 00:01:07,680
Speaker 3: inputting an equities trade actually works like where it kind

19
00:01:07,720 --> 00:01:10,880
Speaker 3: of shows up. So that's a big question. I suspect

20
00:01:10,920 --> 00:01:12,920
Speaker 3: a lot of people, even people who are in markets,

21
00:01:12,959 --> 00:01:16,440
Speaker 3: probably don't know like the entire sequence of events, partly

22
00:01:16,480 --> 00:01:19,280
Speaker 3: because it's gotten more complicated over the years with like

23
00:01:19,400 --> 00:01:22,160
Speaker 3: reg nms do you remember that, Yeah, and stuff like that.

24
00:01:22,640 --> 00:01:26,600
Speaker 3: The other thing I've been realizing about trading, obviously, the

25
00:01:26,640 --> 00:01:29,640
Speaker 3: big trend here is high frequency trading. Yeah, right, and

26
00:01:29,760 --> 00:01:33,679
Speaker 3: it's just getting faster and faster and faster. When we

27
00:01:33,880 --> 00:01:37,240
Speaker 3: first started writing about HFT, I guess in the sort

28
00:01:37,280 --> 00:01:41,600
Speaker 3: of like mid two thousands, after the financial crisis, I

29
00:01:41,640 --> 00:01:45,959
Speaker 3: remember thinking that it was all about the actual algorithm

30
00:01:46,240 --> 00:01:50,720
Speaker 3: and finding like a really smart pattern in financial markets

31
00:01:50,720 --> 00:01:53,400
Speaker 3: to exploit. But the more I learn about it and

32
00:01:53,440 --> 00:01:56,000
Speaker 3: the more I read about it, I kind of realized

33
00:01:56,080 --> 00:01:58,920
Speaker 3: it's not really that. It's about exploiting the market structure.

34
00:01:59,200 --> 00:02:02,240
Speaker 2: Yeah. Yeah, And there's so many you know, we had

35
00:02:02,240 --> 00:02:04,560
Speaker 2: that I forget who we were talking to recently, Oh,

36
00:02:04,680 --> 00:02:07,200
Speaker 2: is the guy from Hudson River Trading, And you know,

37
00:02:07,280 --> 00:02:09,919
Speaker 2: there were the famous like wire wars where it's like, no,

38
00:02:10,000 --> 00:02:12,480
Speaker 2: I want to be one inch closer to the main server,

39
00:02:12,680 --> 00:02:14,480
Speaker 2: et cetera. It's like, God, this is like a good

40
00:02:14,560 --> 00:02:16,880
Speaker 2: use of brain power to like we're going to solve

41
00:02:16,919 --> 00:02:17,519
Speaker 2: the market.

42
00:02:17,320 --> 00:02:19,639
Speaker 3: Because the one more nanosecond faster.

43
00:02:19,680 --> 00:02:22,440
Speaker 2: Nano second faster, et cetera. But you know the other

44
00:02:22,480 --> 00:02:24,680
Speaker 2: thing you know, I sort of related to this. One

45
00:02:24,680 --> 00:02:27,240
Speaker 2: of my long standing questions is, you know, a jobs

46
00:02:27,240 --> 00:02:29,839
Speaker 2: report will drop at eight thirty on a Friday and

47
00:02:30,160 --> 00:02:33,359
Speaker 2: the market immediately moves, And I'm like, how did that happen?

48
00:02:33,440 --> 00:02:36,240
Speaker 2: Because it didn't happen because someone was staring they had

49
00:02:36,240 --> 00:02:39,280
Speaker 2: their like fingers above the buyer the sell button, but

50
00:02:39,400 --> 00:02:42,080
Speaker 2: also had something had to be programmed such that that

51
00:02:42,200 --> 00:02:45,760
Speaker 2: data could instantly be ingested and then some sort of

52
00:02:45,800 --> 00:02:48,560
Speaker 2: like directional trade was made based on that. But I

53
00:02:48,600 --> 00:02:50,440
Speaker 2: don't know how that happens. I don't know how that works.

54
00:02:50,520 --> 00:02:52,840
Speaker 3: There's also, of course, the overall question of what all

55
00:02:52,880 --> 00:02:56,320
Speaker 3: this electronic trading actually means for the market itself. Yes,

56
00:02:56,480 --> 00:02:59,240
Speaker 3: and people talk about things, you know, with the multi

57
00:02:59,240 --> 00:03:04,200
Speaker 3: strap fund, getting these feedback loops and maybe increasing volatility

58
00:03:04,200 --> 00:03:06,400
Speaker 3: in the market and things like that, So we should discuss.

59
00:03:06,560 --> 00:03:08,640
Speaker 2: And I just add one more thing. You said, what

60
00:03:08,639 --> 00:03:10,560
Speaker 2: does it mean for the market itself? What does it

61
00:03:10,600 --> 00:03:12,920
Speaker 2: mean for society it self that so much effort is

62
00:03:13,000 --> 00:03:16,760
Speaker 2: being placed on like getting a meter closer to the

63
00:03:16,800 --> 00:03:19,359
Speaker 2: server room or whatever, And why is this a good

64
00:03:19,480 --> 00:03:23,079
Speaker 2: use of time? And has this improved capital allocation? I'm

65
00:03:23,120 --> 00:03:25,200
Speaker 2: really excited to say we do. In fact, have I

66
00:03:25,200 --> 00:03:28,320
Speaker 2: think truly the perfect guest to talk about this, someone

67
00:03:28,320 --> 00:03:31,320
Speaker 2: who has a sort of an extraordinary body of life's

68
00:03:31,360 --> 00:03:33,640
Speaker 2: work in a range of areas that is very distinct

69
00:03:33,720 --> 00:03:37,080
Speaker 2: from almost any other academic or researcher that I can

70
00:03:37,240 --> 00:03:39,800
Speaker 2: think of. We're going to be speaking with Donald McKenzie.

71
00:03:39,800 --> 00:03:42,440
Speaker 2: He's a professor of sociology at the University of Edinburgh

72
00:03:42,520 --> 00:03:45,280
Speaker 2: in Scotland. I first came across his work. He wrote

73
00:03:45,320 --> 00:03:48,520
Speaker 2: a fantastic book called An Engine Not a Camera, which

74
00:03:48,600 --> 00:03:51,360
Speaker 2: is sort of about finance and the original birth of

75
00:03:51,440 --> 00:03:55,480
Speaker 2: quantitative finance and the use of advanced models, and how

76
00:03:55,600 --> 00:03:58,160
Speaker 2: these models didn't just reflect what was going on in

77
00:03:58,200 --> 00:04:00,480
Speaker 2: the real world, but how the adoption of these models

78
00:04:00,480 --> 00:04:03,920
Speaker 2: then created this feedback loop, the engine effect, such that

79
00:04:04,000 --> 00:04:08,720
Speaker 2: it actually started to drive markets themselves. More recently he

80
00:04:08,760 --> 00:04:10,800
Speaker 2: wrote a book called Trading at the Speed of Light,

81
00:04:10,920 --> 00:04:14,320
Speaker 2: all about high frequency trading. It's also written recently a

82
00:04:14,360 --> 00:04:18,719
Speaker 2: book about digital advertising, and so truly a polymath in

83
00:04:18,760 --> 00:04:21,920
Speaker 2: the world of thinking about this relationship between industry and

84
00:04:21,960 --> 00:04:26,360
Speaker 2: the sort of technological substrates that drive them. Professor McKenzie,

85
00:04:26,360 --> 00:04:28,640
Speaker 2: thank you so much for coming on outlaws.

86
00:04:28,720 --> 00:04:30,880
Speaker 4: Well, thank you very much for inviting me to do that.

87
00:04:31,360 --> 00:04:34,839
Speaker 2: Absolutely, what do you start off by telling us the

88
00:04:34,920 --> 00:04:38,520
Speaker 2: distalt of your life work? What is your core underlying

89
00:04:38,720 --> 00:04:42,599
Speaker 2: interest such that it has produced books in these various realms?

90
00:04:42,680 --> 00:04:46,599
Speaker 4: Yeah, I mean fundamentally, I'm a sociologist of technology, so

91
00:04:47,400 --> 00:04:54,680
Speaker 4: I'm interested in the technical systems that effect or could

92
00:04:54,760 --> 00:04:59,400
Speaker 4: effect all of us, you know, so over time, first

93
00:04:59,600 --> 00:05:05,320
Speaker 4: major project in that area was on nuclear missile guidance technology.

94
00:05:05,640 --> 00:05:11,720
Speaker 4: Then I moved on to safety critical computing technology, then

95
00:05:11,760 --> 00:05:16,640
Speaker 4: the work on financial models that you've just mentioned, then

96
00:05:16,720 --> 00:05:22,239
Speaker 4: high frequency trading, them digital advertising, you know, because as

97
00:05:22,279 --> 00:05:25,640
Speaker 4: well as driving us all insane by our ads that

98
00:05:25,680 --> 00:05:28,520
Speaker 4: we don't want to see appearing on our screens, that's

99
00:05:28,560 --> 00:05:33,120
Speaker 4: also of course the big funding source for much of

100
00:05:33,160 --> 00:05:35,840
Speaker 4: the everyday digital world. And then most recently of all,

101
00:05:35,880 --> 00:05:40,040
Speaker 4: I've started working on AI and large language models. So

102
00:05:40,080 --> 00:05:42,240
Speaker 4: you can see that. You can see the picture. They're

103
00:05:42,240 --> 00:05:45,839
Speaker 4: all highly technical areas one way or the other, the

104
00:05:46,160 --> 00:05:47,440
Speaker 4: all affect all of us.

105
00:05:47,680 --> 00:05:49,440
Speaker 3: The other reason we wanted to talk to you is

106
00:05:49,480 --> 00:05:53,600
Speaker 3: because you come at everything from this sociological perspective, and

107
00:05:53,680 --> 00:05:57,960
Speaker 3: I absolutely love it when like anthropologists and sociologists go

108
00:05:58,000 --> 00:06:01,000
Speaker 3: to Wall Street and write about it. Why did you

109
00:06:01,080 --> 00:06:04,960
Speaker 3: take that approach, especially with your high frequency trading book.

110
00:06:05,440 --> 00:06:11,240
Speaker 4: Yeah, well, I don't do kind of like quantitative social science,

111
00:06:11,560 --> 00:06:13,440
Speaker 4: you know, I'd leave that, for example, as far as

112
00:06:13,560 --> 00:06:16,200
Speaker 4: markets are concerned, I leave that to economists. What I

113
00:06:16,240 --> 00:06:20,440
Speaker 4: do I like talking to people. I like going looking

114
00:06:20,680 --> 00:06:24,000
Speaker 4: at staff to the extent that you can look at it.

115
00:06:24,279 --> 00:06:29,000
Speaker 4: I like tracing how things have developed through time. My

116
00:06:29,480 --> 00:06:31,520
Speaker 4: work's often got us kind of you know, something of

117
00:06:31,520 --> 00:06:35,400
Speaker 4: a historical dimension to it. But the most fun bit

118
00:06:35,760 --> 00:06:39,359
Speaker 4: isn't writing the books. The most fun bit is talking

119
00:06:39,400 --> 00:06:42,240
Speaker 4: to people. And that's a bit I've always enjoyed most.

120
00:06:42,720 --> 00:06:44,880
Speaker 3: One interesting thing in this book, Joe, I don't know

121
00:06:44,880 --> 00:06:47,680
Speaker 3: if you noticed, but Donald writes down like all his

122
00:06:47,920 --> 00:06:51,159
Speaker 3: numbers of sources and who they are, so like, you know,

123
00:06:51,200 --> 00:06:54,440
Speaker 3: people from the exchange, people from high frequency trading firms,

124
00:06:54,760 --> 00:06:57,440
Speaker 3: what their seniority is, which is something I hadn't really

125
00:06:57,520 --> 00:06:58,320
Speaker 3: seen before.

126
00:06:58,520 --> 00:07:01,320
Speaker 2: It is a really cool thing. By the way, as

127
00:07:01,400 --> 00:07:04,159
Speaker 2: Donald says, the fun part is talking to people, not

128
00:07:04,240 --> 00:07:06,760
Speaker 2: so much the writing of the book. As two people

129
00:07:06,839 --> 00:07:09,880
Speaker 2: who talk to people every day and have never written

130
00:07:09,920 --> 00:07:13,040
Speaker 2: a book, I feel already. Now granted we never actually

131
00:07:13,040 --> 00:07:15,040
Speaker 2: went through the process of writing the book because the

132
00:07:15,080 --> 00:07:16,840
Speaker 2: talking part is so much more fun and I don't

133
00:07:16,840 --> 00:07:18,920
Speaker 2: want to ever take a pause from the talking. Yeah,

134
00:07:18,920 --> 00:07:22,000
Speaker 2: so I already feel like, to some extent, Donald is

135
00:07:22,040 --> 00:07:24,680
Speaker 2: a kindred spirit. It's like, it's fun to talk to people,

136
00:07:24,800 --> 00:07:27,120
Speaker 2: isn't it. Talk to us about how you find them?

137
00:07:27,160 --> 00:07:30,160
Speaker 2: You know, it's like, okay, HFT is interesting to you,

138
00:07:30,200 --> 00:07:32,360
Speaker 2: and like you just want to have some conversations. What

139
00:07:32,520 --> 00:07:36,920
Speaker 2: is the sociologist's toolkit here for knowing who to talk to?

140
00:07:37,720 --> 00:07:41,920
Speaker 4: Yeah, well, it's always difficult, and it's always very ad hoc.

141
00:07:42,000 --> 00:07:46,360
Speaker 4: There's always a lot of luck involved in it. And

142
00:07:46,760 --> 00:07:51,720
Speaker 4: with a financial market topic, I will typically start reading

143
00:07:51,720 --> 00:07:56,720
Speaker 4: the Financial Times, finding names in the Financial Times, approaching

144
00:07:57,480 --> 00:08:00,280
Speaker 4: those people, and then maybe they pass me on to

145
00:08:00,440 --> 00:08:04,280
Speaker 4: other people. But you know, there's also, as I said,

146
00:08:04,320 --> 00:08:07,600
Speaker 4: dumb luck involved. Like a crucial moment in the work

147
00:08:07,640 --> 00:08:09,920
Speaker 4: I did in high frequency trading was going to interview

148
00:08:09,960 --> 00:08:12,960
Speaker 4: someone at the start of it, and framed on his

149
00:08:13,120 --> 00:08:17,600
Speaker 4: wall was the front cover of an issue of Forbes

150
00:08:18,160 --> 00:08:22,560
Speaker 4: with the headline of the article free Enterprise comes to

151
00:08:22,640 --> 00:08:26,840
Speaker 4: Wall Street. And they thought, oh, that sounds kind of interesting,

152
00:08:27,480 --> 00:08:29,320
Speaker 4: and I checked that out, and it was to do

153
00:08:29,440 --> 00:08:35,520
Speaker 4: with a new electronic stock exchange called Island. And it

154
00:08:35,600 --> 00:08:39,160
Speaker 4: turned out the Island, the story of Island was completely

155
00:08:39,200 --> 00:08:42,400
Speaker 4: interwoven with the story of high frequency trading.

156
00:08:43,440 --> 00:08:46,520
Speaker 3: Before we get into exactly what hygh frequency trading is

157
00:08:46,559 --> 00:08:50,600
Speaker 3: and how it fits into placing orders for equities or

158
00:08:50,840 --> 00:08:54,280
Speaker 3: futures or bonds, I have a cultural question, which is,

159
00:08:54,320 --> 00:08:58,480
Speaker 3: whenever you go into an HFT firm's office, it always

160
00:08:58,520 --> 00:09:00,280
Speaker 3: looks like a tech company.

161
00:09:00,440 --> 00:09:01,840
Speaker 2: Yeah, right, it's very just.

162
00:09:04,720 --> 00:09:07,480
Speaker 3: And very modern, Like why do you think they've taken

163
00:09:07,559 --> 00:09:11,400
Speaker 3: that approach? How did that esthetic become the norm?

164
00:09:11,840 --> 00:09:16,520
Speaker 4: Yeah, that's a really good indicator of cultural change, because,

165
00:09:16,559 --> 00:09:20,520
Speaker 4: of course, previous to that, the sort of dominant image

166
00:09:20,520 --> 00:09:22,880
Speaker 4: we might have of a financial market would be the

167
00:09:22,920 --> 00:09:27,360
Speaker 4: trading floor of the New York Stock Exchange. You know,

168
00:09:27,520 --> 00:09:33,160
Speaker 4: folks and colored jackets. They're typically televised when even nowadays,

169
00:09:33,160 --> 00:09:35,240
Speaker 4: when you know, so there's been a big drop in

170
00:09:35,280 --> 00:09:38,040
Speaker 4: the market or something, so the cameras try to catch

171
00:09:38,040 --> 00:09:43,720
Speaker 4: somebody who's looking kind of glom and and worried. So

172
00:09:43,800 --> 00:09:46,120
Speaker 4: we think of that as what finance is.

173
00:09:46,320 --> 00:09:49,240
Speaker 2: Or we think about like the bud Fox and Wall

174
00:09:49,280 --> 00:09:52,160
Speaker 2: Street of a bunch of guys and slickbag hair sort

175
00:09:52,160 --> 00:09:56,040
Speaker 2: of you know, on theiler. Yeah, yelling at each other

176
00:09:56,240 --> 00:09:58,480
Speaker 2: looking at a green screen. Sorry, I didn't mean to

177
00:09:58,480 --> 00:10:01,719
Speaker 2: interrupt it, but those are the two things people imagine.

178
00:10:01,880 --> 00:10:04,880
Speaker 4: Yeah, yeah, and you know, there's was a transition involved,

179
00:10:04,880 --> 00:10:09,600
Speaker 4: but by and large, the hyphogency trading firms higher people

180
00:10:09,640 --> 00:10:14,400
Speaker 4: who know how to code often you know, with higher

181
00:10:14,440 --> 00:10:19,040
Speaker 4: degrees and mathematical kinds of subjects, and even the people

182
00:10:19,080 --> 00:10:25,360
Speaker 4: who refer to themselves as traders often have that kind

183
00:10:25,960 --> 00:10:31,120
Speaker 4: of background. You know. I'm sure when the visitor isn't there,

184
00:10:31,160 --> 00:10:33,760
Speaker 4: there'll be a fair bit of swearing at the screen

185
00:10:33,920 --> 00:10:36,960
Speaker 4: when something goes wrong and and that kind of stuff.

186
00:10:36,960 --> 00:10:40,720
Speaker 4: But you're right that the normal experience of those trading rooms,

187
00:10:41,000 --> 00:10:46,120
Speaker 4: they're quiet there orderly, and you could indeed make mistake

188
00:10:46,200 --> 00:10:48,120
Speaker 4: them for a Silicon Valley startup.

189
00:10:48,400 --> 00:10:52,240
Speaker 2: Yeah, and you see people in jeans, And I visited

190
00:10:52,280 --> 00:10:54,040
Speaker 2: one and like, I think I saw the CEO and

191
00:10:54,080 --> 00:10:56,600
Speaker 2: he was just like wearing as like college T shirt

192
00:10:57,000 --> 00:10:58,520
Speaker 2: or something like that.

193
00:10:58,559 --> 00:11:01,319
Speaker 4: With strong memory I've got because then in the previous

194
00:11:01,360 --> 00:11:03,679
Speaker 4: work they worked for an engine, not a camera, and

195
00:11:03,720 --> 00:11:06,520
Speaker 4: some follow on stuff. I would often go to investment banks,

196
00:11:06,920 --> 00:11:10,240
Speaker 4: and in investment banks, you know, I kind of had

197
00:11:10,240 --> 00:11:13,560
Speaker 4: to wear suit and a nice shirt and a tie

198
00:11:14,040 --> 00:11:16,240
Speaker 4: and so on. So when I started interviewing in high

199
00:11:16,280 --> 00:11:19,720
Speaker 4: frequency trading, I turned up at one firm dressed like that,

200
00:11:20,000 --> 00:11:23,920
Speaker 4: and the owner of the firms sort of snarled at me,

201
00:11:24,360 --> 00:11:25,439
Speaker 4: you're over dressed.

202
00:11:26,200 --> 00:11:29,640
Speaker 2: Wow, you mentioned island. What do you tell us that story?

203
00:11:29,720 --> 00:11:31,840
Speaker 2: You know, I want to get more into the tech, etc.

204
00:11:32,280 --> 00:11:34,920
Speaker 2: But you're like, Okay, this turned out to be an exchange.

205
00:11:35,120 --> 00:11:37,880
Speaker 2: What was distinct? What is island? I've heard of it,

206
00:11:37,920 --> 00:11:39,959
Speaker 2: but that's again one of these things that I've heard

207
00:11:39,960 --> 00:11:42,160
Speaker 2: of it. And then I moved on what was distinct

208
00:11:42,200 --> 00:11:44,640
Speaker 2: about this and why is it so interwoven into the

209
00:11:44,720 --> 00:11:48,600
Speaker 2: history of HFT. How is it different from other exchanges

210
00:11:48,640 --> 00:11:50,320
Speaker 2: that have existed for hundreds of years.

211
00:11:50,440 --> 00:11:52,800
Speaker 4: Yeah, yeah, you know, I'm going to oversimplify, of course,

212
00:11:52,840 --> 00:11:56,040
Speaker 4: because there were predecessors to islands. You know, we're a

213
00:11:56,120 --> 00:11:57,839
Speaker 4: little bit like it and so on, but that would

214
00:11:57,880 --> 00:12:00,560
Speaker 4: take us too long to go into. I mean mentally,

215
00:12:00,960 --> 00:12:06,840
Speaker 4: trading on Island was organized around an electronic order book,

216
00:12:07,120 --> 00:12:11,240
Speaker 4: which was is a list all the bids to buy

217
00:12:11,480 --> 00:12:15,199
Speaker 4: our offers to sell the shares in question, and that

218
00:12:15,320 --> 00:12:19,840
Speaker 4: electronic order book is managed by something called a matching engine,

219
00:12:19,880 --> 00:12:24,160
Speaker 4: and as the name implies, that looks for a match,

220
00:12:24,200 --> 00:12:26,360
Speaker 4: in other words, a bid to buy and an offer

221
00:12:26,400 --> 00:12:29,520
Speaker 4: to sell at the same price, and when it finds

222
00:12:29,600 --> 00:12:35,200
Speaker 4: that couple, it consummates the trade and the trade is done.

223
00:12:35,400 --> 00:12:41,600
Speaker 4: So it's all that electronically. There's no direct human negotiation involved.

224
00:12:41,720 --> 00:12:45,760
Speaker 4: You just enter your orders into the order book and

225
00:12:45,880 --> 00:12:49,320
Speaker 4: the matching engine either executes them or fails to find

226
00:12:49,320 --> 00:12:54,040
Speaker 4: a match. There were exchanges prior to Island that worked

227
00:12:54,080 --> 00:12:57,640
Speaker 4: in that kind of way, but what was distinctive about

228
00:12:57,720 --> 00:13:03,120
Speaker 4: Island is that it's matching engine was blisteringly fast by

229
00:13:03,280 --> 00:13:05,800
Speaker 4: the standards of the day, which were you know, it

230
00:13:05,920 --> 00:13:10,560
Speaker 4: was essentially the late nineteen nineties. So the closest analog

231
00:13:11,000 --> 00:13:15,640
Speaker 4: was a system called Instanet, and it might take a

232
00:13:15,679 --> 00:13:19,000
Speaker 4: couple of seconds the matching engine to find the match

233
00:13:19,160 --> 00:13:22,240
Speaker 4: and execute the trade, and of course, for a human

234
00:13:22,280 --> 00:13:26,600
Speaker 4: being sitting there, even if they're in patient, two seconds

235
00:13:26,720 --> 00:13:30,280
Speaker 4: is not a very long time. Island improved on that

236
00:13:30,320 --> 00:13:36,760
Speaker 4: one thousandfold, so it could execute trades in two milliseconds.

237
00:13:36,800 --> 00:13:41,720
Speaker 4: Two thousandths of a second. So that was the opening

238
00:13:42,160 --> 00:13:48,160
Speaker 4: for high frequency trading. That with exchange, I mean strictly,

239
00:13:48,360 --> 00:13:50,640
Speaker 4: Island was not an exchange. It was what was called

240
00:13:50,640 --> 00:13:54,400
Speaker 4: an electronic communications network or ECND. But I'll cause an

241
00:13:54,440 --> 00:13:58,160
Speaker 4: exchange for simplicity. If you've got an exchange like that

242
00:13:59,000 --> 00:14:04,199
Speaker 4: and you've got an automated trading system, it's a marriage

243
00:14:04,200 --> 00:14:07,280
Speaker 4: made in heaven. The two things, the exchange and the

244
00:14:07,320 --> 00:14:15,400
Speaker 4: trading firm, fit each other very very well. And amongst

245
00:14:15,440 --> 00:14:20,080
Speaker 4: the consequences of that is that liquidity. In Ireland, it

246
00:14:20,160 --> 00:14:22,960
Speaker 4: traded Nasdaq stocks, and this is the time of the

247
00:14:23,000 --> 00:14:25,040
Speaker 4: dot com bubble, of course, where there's a lot of

248
00:14:25,080 --> 00:14:29,600
Speaker 4: trading of NASDAK tech stocks. Island brought a lot of

249
00:14:29,680 --> 00:14:34,720
Speaker 4: liquidity to that market. So that's you get a kind

250
00:14:34,760 --> 00:14:41,280
Speaker 4: of feedback loop where you get automated trading bringing liquidity

251
00:14:41,680 --> 00:14:45,120
Speaker 4: to exchanges that have the kind of technical features that

252
00:14:45,200 --> 00:14:49,400
Speaker 4: make high frequency trading attractive and feasible. So the established

253
00:14:49,440 --> 00:14:53,480
Speaker 4: exchanges started to have to change how they did things

254
00:14:53,640 --> 00:14:55,880
Speaker 4: because otherwise they were going to lose out to the

255
00:14:56,440 --> 00:15:01,760
Speaker 4: new exchanges. And that's basically the feedback loop that's created

256
00:15:01,880 --> 00:15:20,080
Speaker 4: today's electronic markets. Joe.

257
00:15:20,080 --> 00:15:23,040
Speaker 3: Whenever I hear terms like millisecond and nanosecond, I just

258
00:15:23,400 --> 00:15:25,800
Speaker 3: it's so hard to wrap my head around what that. Actually,

259
00:15:25,960 --> 00:15:30,040
Speaker 3: it's faster than that, for sure, Donald.

260
00:15:30,080 --> 00:15:32,880
Speaker 4: I can actually help there. And I'm going to hold

261
00:15:32,920 --> 00:15:38,440
Speaker 4: my fingers. I'm holding them thirty centimeters apart, or since

262
00:15:38,440 --> 00:15:42,920
Speaker 4: you're in the US, i'll say one foot apart. At

263
00:15:42,920 --> 00:15:45,920
Speaker 4: the speed of light in the vacuum, it takes a

264
00:15:46,040 --> 00:15:51,200
Speaker 4: nanosecond to get from one finger to the other finger.

265
00:15:51,680 --> 00:15:58,240
Speaker 4: And that's an indication of how fast automated trading, specifically

266
00:15:58,320 --> 00:16:02,880
Speaker 4: high frequency trading, has been. That when I started working

267
00:16:03,160 --> 00:16:07,280
Speaker 4: on the topic in roughly twenty eleven, people were still

268
00:16:07,320 --> 00:16:11,000
Speaker 4: talking about milliseconds or thousands of a second. Two or

269
00:16:11,000 --> 00:16:14,760
Speaker 4: three years later it had become microseconds or millions of

270
00:16:14,760 --> 00:16:17,040
Speaker 4: a second. And by the time that I was finishing

271
00:16:17,080 --> 00:16:24,320
Speaker 4: the research, nanoseconds were starting to account light traveling that

272
00:16:24,640 --> 00:16:28,920
Speaker 4: thirty centimeters traveling that food that matter to ewing see

273
00:16:28,960 --> 00:16:33,880
Speaker 4: trading by roughly twenty eighteen, twenty nineteen, twenty twenty round about.

274
00:16:33,920 --> 00:16:36,920
Speaker 3: Then that's very helpful. I do have questions about the

275
00:16:36,920 --> 00:16:39,960
Speaker 3: physical realities of how fast we can actually go with

276
00:16:40,040 --> 00:16:42,760
Speaker 3: all this stuff. But before we go any further, can

277
00:16:42,800 --> 00:16:46,080
Speaker 3: you talk about the process of let's just focus on

278
00:16:46,120 --> 00:16:50,640
Speaker 3: the equity market for now. Someone places an order to

279
00:16:50,760 --> 00:16:54,720
Speaker 3: buy or sell a stock, what actually happens in the

280
00:16:54,760 --> 00:16:59,360
Speaker 3: ecosystem between traders and market makers and the exchange that

281
00:16:59,480 --> 00:17:01,920
Speaker 3: makes that happen, And what does it mean to actually

282
00:17:01,960 --> 00:17:04,200
Speaker 3: make that happen and execute the trade.

283
00:17:04,440 --> 00:17:08,880
Speaker 4: So what happens is your order via the broker your

284
00:17:09,000 --> 00:17:11,400
Speaker 4: use I mean, and the brokerage system is no longer

285
00:17:11,400 --> 00:17:15,560
Speaker 4: a human being via the brokerage system, gets placed in

286
00:17:15,640 --> 00:17:22,800
Speaker 4: the exchange's order book, and then one of two things happens.

287
00:17:22,880 --> 00:17:28,840
Speaker 4: The first is if the matching engine can find an

288
00:17:28,920 --> 00:17:32,840
Speaker 4: existing order in the order book that matches the price

289
00:17:33,160 --> 00:17:37,399
Speaker 4: of your order, it executes the trade, and the trade

290
00:17:37,880 --> 00:17:42,640
Speaker 4: then happens not quite instantly, but you know, very very

291
00:17:42,800 --> 00:17:47,080
Speaker 4: very fast, and you're done. That's it. It's over. If,

292
00:17:47,119 --> 00:17:49,960
Speaker 4: on the other hand, there is no match as the

293
00:17:50,080 --> 00:17:55,920
Speaker 4: order book stands, your order rests in the order book

294
00:17:56,640 --> 00:18:00,760
Speaker 4: and it stays there until either you cancel it or

295
00:18:01,200 --> 00:18:06,280
Speaker 4: a matching order comes along, and then it's executed at

296
00:18:06,359 --> 00:18:09,320
Speaker 4: that point. So that's the basic process, you know.

297
00:18:09,600 --> 00:18:12,959
Speaker 2: So it occurs to me like gains of speed in

298
00:18:13,080 --> 00:18:17,560
Speaker 2: trading have been happening forever, long before we were talking about,

299
00:18:17,640 --> 00:18:21,680
Speaker 2: you know, anything electronic. I'm sure, Yeah, other technologies exist.

300
00:18:21,880 --> 00:18:25,680
Speaker 2: Technological evolution is a long time thing. Is pretty banal statement,

301
00:18:25,960 --> 00:18:28,920
Speaker 2: I suppose. But what I'm curious about is the sense

302
00:18:28,960 --> 00:18:32,000
Speaker 2: of which a change in degree becomes a change in

303
00:18:32,160 --> 00:18:36,399
Speaker 2: kind essentially. Yeah, so that like when you go from

304
00:18:36,640 --> 00:18:40,840
Speaker 2: one second to a thousandth of a second to a nanosecond,

305
00:18:41,119 --> 00:18:44,760
Speaker 2: how does that change, say, like the types of strategies

306
00:18:44,800 --> 00:18:47,480
Speaker 2: that can then be employed, or the types of skills

307
00:18:47,520 --> 00:18:49,879
Speaker 2: that might be required to be a successful trader in

308
00:18:49,920 --> 00:18:53,760
Speaker 2: the nanosecond era versus the one second era. Like talk

309
00:18:53,800 --> 00:18:55,480
Speaker 2: to us about like that relationship.

310
00:18:55,600 --> 00:18:58,000
Speaker 4: Yeah, yeah, Well, there's a wonderful book by the historian

311
00:18:58,000 --> 00:19:01,320
Speaker 4: the technology Jimena Canalis, which is called a tenth of

312
00:19:01,359 --> 00:19:06,080
Speaker 4: a second a history. And the significance of the tenth

313
00:19:06,160 --> 00:19:12,320
Speaker 4: of a second is that the generally accepted lower threshold

314
00:19:12,680 --> 00:19:17,879
Speaker 4: of the human perceptibility of time. You know, basically, we

315
00:19:18,000 --> 00:19:23,960
Speaker 4: just can't mentally process time intervals that are less.

316
00:19:24,080 --> 00:19:27,560
Speaker 2: So Tracy, don't feel bad about not being an intuition

317
00:19:27,680 --> 00:19:29,840
Speaker 2: for a nanosecond less.

318
00:19:29,640 --> 00:19:32,480
Speaker 4: Than the tenth of a second, you know. So what

319
00:19:32,680 --> 00:19:39,359
Speaker 4: essentially happened is that we've moved from that kind of

320
00:19:39,440 --> 00:19:41,960
Speaker 4: you know, the tenth of a second or longer, from

321
00:19:42,280 --> 00:19:48,560
Speaker 4: from that kind of epoch into an epoch where human beings,

322
00:19:48,560 --> 00:19:51,840
Speaker 4: I mean, they can still be an overall control of

323
00:19:52,080 --> 00:19:56,680
Speaker 4: the system, but they can't actually execute the actual trading

324
00:19:57,119 --> 00:20:02,680
Speaker 4: decisions fast enough not to be outrun by an algorithm.

325
00:20:02,720 --> 00:20:07,639
Speaker 4: So we moved from a kind of human centered form

326
00:20:08,280 --> 00:20:12,880
Speaker 4: of trading to a machine centered form of trading. And

327
00:20:12,920 --> 00:20:15,840
Speaker 4: the you know, the actual threshold of the change is

328
00:20:15,880 --> 00:20:19,200
Speaker 4: probably around that tenth of the second amount.

329
00:20:19,640 --> 00:20:23,040
Speaker 3: I have another cultural and I guess market structure question.

330
00:20:23,119 --> 00:20:25,240
Speaker 3: But one thing that I always thought was interesting about

331
00:20:25,280 --> 00:20:29,760
Speaker 3: high frequency trading was that the banks didn't really get

332
00:20:29,760 --> 00:20:33,240
Speaker 3: into it, which, you know, there's one big reason why,

333
00:20:33,280 --> 00:20:35,879
Speaker 3: which is the ban on prop trading after two thousand

334
00:20:35,880 --> 00:20:39,400
Speaker 3: and eights, but even before then, they just never seem

335
00:20:39,480 --> 00:20:42,560
Speaker 3: to be able to compete with independent firms. Why did

336
00:20:42,560 --> 00:20:43,080
Speaker 3: that happen?

337
00:20:43,720 --> 00:20:47,040
Speaker 4: Yeah, it's I've asked people that, and there's I think

338
00:20:47,160 --> 00:20:51,080
Speaker 4: complex sorts of reasons. And let's let it be said,

339
00:20:51,119 --> 00:20:54,000
Speaker 4: some banks have been more successful than others. Banks are

340
00:20:54,000 --> 00:20:59,760
Speaker 4: not always bad, though, most banks are bad at it.

341
00:21:00,320 --> 00:21:03,440
Speaker 4: One way of thinking about is that typically a bank

342
00:21:03,640 --> 00:21:08,640
Speaker 4: will have an IT department that's separate from the other

343
00:21:08,800 --> 00:21:14,159
Speaker 4: functions of the bank, like trading, like market making and

344
00:21:14,200 --> 00:21:16,160
Speaker 4: so on. So if you know, if you're a trader,

345
00:21:16,760 --> 00:21:22,040
Speaker 4: you got to persuade the IT people to give you

346
00:21:22,240 --> 00:21:25,520
Speaker 4: a fast enough system, which involves them, you know, maybe

347
00:21:25,520 --> 00:21:30,080
Speaker 4: writing some new software, buying some new kit. So you

348
00:21:30,160 --> 00:21:34,040
Speaker 4: need to get higher level management sign off on it,

349
00:21:34,400 --> 00:21:38,760
Speaker 4: and it all takes time. Whereas the high things trading

350
00:21:38,800 --> 00:21:42,920
Speaker 4: firms are typically pretty small. You know, fifty people is

351
00:21:43,440 --> 00:21:46,959
Speaker 4: a decent size firm. One hundred and fifty people is

352
00:21:47,359 --> 00:21:51,320
Speaker 4: you know, it is a reasonably big high frequency trading firm.

353
00:21:51,960 --> 00:21:57,119
Speaker 4: Very often those firms are owned and run by this

354
00:21:57,240 --> 00:21:59,560
Speaker 4: you know, the people who founded them, so that there

355
00:21:59,600 --> 00:22:01,480
Speaker 4: is a ball us or bosses. But you know, other

356
00:22:01,560 --> 00:22:06,000
Speaker 4: than that, it is a relatively flat organizational structure. If,

357
00:22:06,000 --> 00:22:08,359
Speaker 4: for example, at least this was the case in the

358
00:22:08,359 --> 00:22:10,280
Speaker 4: early days of if we can see trading, it's not

359
00:22:10,359 --> 00:22:11,960
Speaker 4: quite as simple as this not, but in the early

360
00:22:12,000 --> 00:22:14,080
Speaker 4: day is igh we can be trading. You know, if

361
00:22:14,520 --> 00:22:20,280
Speaker 4: some IT firm came out with a new, better, faster

362
00:22:21,280 --> 00:22:24,399
Speaker 4: server and you were a trader in a firm like that,

363
00:22:24,480 --> 00:22:27,399
Speaker 4: and you know, okay, well let's get this new server.

364
00:22:28,480 --> 00:22:31,320
Speaker 4: You could just use your own personal credit card to

365
00:22:31,800 --> 00:22:35,360
Speaker 4: buy the server, get it delivered to your office, and

366
00:22:35,720 --> 00:22:38,760
Speaker 4: then get your engineers to take it out to whatever

367
00:22:38,840 --> 00:22:41,720
Speaker 4: data center that they were trading in, and get it

368
00:22:41,760 --> 00:22:45,359
Speaker 4: installed straight away. And you know that could maybe that

369
00:22:45,400 --> 00:22:48,880
Speaker 4: would be a week or ten days or something worse.

370
00:22:49,080 --> 00:22:51,159
Speaker 4: In the bank, you'd be doing pretty well if you

371
00:22:51,160 --> 00:22:52,760
Speaker 4: could achieve that within six months.

372
00:22:53,119 --> 00:22:55,840
Speaker 3: Yeah, that's amazing.

373
00:22:55,920 --> 00:22:56,359
Speaker 4: We don't know.

374
00:22:56,960 --> 00:23:00,360
Speaker 2: We don't know anything about long waits for computers.

375
00:23:01,040 --> 00:23:05,040
Speaker 3: No, we surely don't. That's sarcasm, by the way. Actually

376
00:23:05,040 --> 00:23:07,359
Speaker 3: this reminds me of something that I wanted to ask,

377
00:23:07,400 --> 00:23:11,840
Speaker 3: which is, we know there's competition between firms high frequency

378
00:23:11,880 --> 00:23:17,880
Speaker 3: traders for the fastest connections to exchanges and things like that.

379
00:23:18,520 --> 00:23:22,720
Speaker 3: There's also competition, I imagine within the firm itself because

380
00:23:23,280 --> 00:23:26,640
Speaker 3: setting aside the credit card anecdote, these can be expensive

381
00:23:26,840 --> 00:23:30,120
Speaker 3: and there's also limited supply. You know, only so many

382
00:23:30,160 --> 00:23:32,639
Speaker 3: servers can be co located where they want to be.

383
00:23:33,560 --> 00:23:37,000
Speaker 3: In your research, how did you actually find executives at

384
00:23:37,200 --> 00:23:41,320
Speaker 3: HFT places? How did they actually allocate the fastest connections

385
00:23:41,359 --> 00:23:42,160
Speaker 3: to which team.

386
00:23:42,400 --> 00:23:45,439
Speaker 4: Ye, what I found was a high frequency trading firms

387
00:23:46,040 --> 00:23:51,280
Speaker 4: fell into two different camps, so to speak. In some

388
00:23:51,359 --> 00:23:57,480
Speaker 4: of them, there were separate trading teams that didn't really

389
00:23:57,520 --> 00:24:01,639
Speaker 4: communicate with each other, and indeed by design didn't communicate

390
00:24:02,119 --> 00:24:05,159
Speaker 4: with each other. In some cases, the office was actually

391
00:24:05,240 --> 00:24:08,520
Speaker 4: laid out in such a way that somebody in one

392
00:24:09,359 --> 00:24:13,879
Speaker 4: team was not very likely accidentally to overhear something said

393
00:24:14,040 --> 00:24:18,359
Speaker 4: by someone in another term. And in those in those firms, yes,

394
00:24:18,480 --> 00:24:22,440
Speaker 4: I mean they are essentially in competition. And I think

395
00:24:22,480 --> 00:24:26,720
Speaker 4: that in that kind of firm, then the results of

396
00:24:26,760 --> 00:24:30,200
Speaker 4: each trading desk, the P and L that profit and loss,

397
00:24:30,600 --> 00:24:33,080
Speaker 4: the little trading teams that are doing best would get

398
00:24:33,480 --> 00:24:38,640
Speaker 4: the available bandwidth on the microwave links that are crucial

399
00:24:38,640 --> 00:24:41,280
Speaker 4: to hyph wingsy trading, and and so on, and maybe

400
00:24:41,320 --> 00:24:44,320
Speaker 4: they would get the fastest machines first, and and so

401
00:24:44,400 --> 00:24:50,080
Speaker 4: on so forth. The other kind of trading firm wasn't

402
00:24:50,240 --> 00:24:56,000
Speaker 4: is operated as a unified entity in some cases even

403
00:24:56,040 --> 00:24:59,440
Speaker 4: without individual profit and loss in the individual p and

404
00:24:59,520 --> 00:25:04,240
Speaker 4: L account for traders. And there's a lot of shared

405
00:25:04,280 --> 00:25:07,400
Speaker 4: infrastructure in that kind of farm, and they died. There's

406
00:25:07,440 --> 00:25:11,120
Speaker 4: also shared infrastructure in the you know, the segregated kind

407
00:25:11,119 --> 00:25:13,280
Speaker 4: of trading firm, because if you're the boss of such

408
00:25:13,280 --> 00:25:17,880
Speaker 4: a firm, there's obviously simple economies in not having completely

409
00:25:17,960 --> 00:25:22,400
Speaker 4: separate IT systems for each each trading for each each

410
00:25:22,520 --> 00:25:25,919
Speaker 4: trading team. But there's that kind of divide. Does the

411
00:25:25,960 --> 00:25:30,280
Speaker 4: firm operate as a unified entity or does the firm

412
00:25:30,600 --> 00:25:35,400
Speaker 4: operate as a sort of aggregate of competing trading teams.

413
00:25:35,800 --> 00:25:39,120
Speaker 2: So actually, let's just you know, on the subject of

414
00:25:39,440 --> 00:25:41,840
Speaker 2: who is the closest or who gets to have their

415
00:25:41,880 --> 00:25:44,400
Speaker 2: server located where? Tell us a little bit more about

416
00:25:44,440 --> 00:25:49,320
Speaker 2: the timeline. So Island emerges in the late nineties, When

417
00:25:49,440 --> 00:25:53,120
Speaker 2: did it start to dawn on people in the trading

418
00:25:53,200 --> 00:25:58,080
Speaker 2: industry that given this new physical reality, given the speed,

419
00:25:58,480 --> 00:26:01,000
Speaker 2: we need to start thinking about out who is going

420
00:26:01,080 --> 00:26:03,760
Speaker 2: to have co location WI do you just start thinking

421
00:26:03,800 --> 00:26:07,480
Speaker 2: about sort of like microwave radio line of sight? Where

422
00:26:07,480 --> 00:26:10,840
Speaker 2: did that speed war? What was the genesis of it?

423
00:26:11,200 --> 00:26:13,359
Speaker 4: Yeah? Yeah, I mean a kind of crucial date was

424
00:26:13,600 --> 00:26:18,359
Speaker 4: two thousand and five where Ireland, which had already been

425
00:26:18,400 --> 00:26:24,119
Speaker 4: bought by Instantet, was acquired by Nasdaq and the New

426
00:26:24,200 --> 00:26:31,240
Speaker 4: York Stock Exchange acquired another of the pioneering electronic trading exchanges.

427
00:26:31,480 --> 00:26:33,040
Speaker 4: This one was called Archipelago.

428
00:26:33,520 --> 00:26:37,160
Speaker 2: And is that just a coincidence that Island and Archipelago, that's.

429
00:26:37,119 --> 00:26:39,120
Speaker 4: Kind of yeah, I'm sure it's not a coincidence. Yeah,

430
00:26:39,520 --> 00:26:44,840
Speaker 4: And they technologically reorganized themselves in New York Stock Exchange,

431
00:26:44,840 --> 00:26:53,560
Speaker 4: and that technologically reorganized themselves around this new insurgent technological

432
00:26:53,960 --> 00:26:57,840
Speaker 4: approach to trading, so to speak. And two thousand and five,

433
00:26:58,240 --> 00:27:00,800
Speaker 4: because of the acquisitions in that year, is a kind

434
00:27:00,840 --> 00:27:05,479
Speaker 4: of noteworthy year. But even before two thousand and five,

435
00:27:06,200 --> 00:27:11,439
Speaker 4: people in trading firms started to become aware that you

436
00:27:11,440 --> 00:27:14,439
Speaker 4: couldn't just do automated trading with the machine sitting in

437
00:27:14,480 --> 00:27:18,240
Speaker 4: your office. For example, there's a Kansas City trading firm

438
00:27:18,280 --> 00:27:21,480
Speaker 4: called trade Bot whose own their dad, Cummings, has written

439
00:27:21,920 --> 00:27:25,640
Speaker 4: really rather nice autobiography. And one of the things that

440
00:27:26,600 --> 00:27:32,800
Speaker 4: Commings came to realize is that trading an island while

441
00:27:32,840 --> 00:27:37,400
Speaker 4: having your machines in Kansas City was placing him at

442
00:27:37,440 --> 00:27:43,040
Speaker 4: a disadvantage. So firms like that started to move their

443
00:27:43,119 --> 00:27:48,240
Speaker 4: machines either directly into the island computer room, or they

444
00:27:48,280 --> 00:27:52,560
Speaker 4: couldn't do that into the offices of another firm in

445
00:27:52,600 --> 00:27:56,080
Speaker 4: the building to shorten the distance, and that kind of

446
00:27:56,080 --> 00:27:58,560
Speaker 4: thing was already in place by two thousand and five.

447
00:27:59,200 --> 00:28:02,600
Speaker 4: Two things then have happened. For a period, there was

448
00:28:02,640 --> 00:28:05,800
Speaker 4: a kind of wild West, so to speak, where there

449
00:28:05,840 --> 00:28:10,160
Speaker 4: are lots of stories of high frequency traders like drilling

450
00:28:10,280 --> 00:28:14,320
Speaker 4: holes in walls so as to shorten the distance between

451
00:28:14,960 --> 00:28:21,360
Speaker 4: their servers and the exchanges matching engines. That is, by

452
00:28:21,400 --> 00:28:23,760
Speaker 4: and large general come to an end. What happens now

453
00:28:24,400 --> 00:28:28,080
Speaker 4: in the data centers of all the major exchanges is

454
00:28:28,560 --> 00:28:32,040
Speaker 4: there is a rule about equal cable length. So if

455
00:28:32,080 --> 00:28:35,159
Speaker 4: you happen to have if a trading firm happens to

456
00:28:35,160 --> 00:28:39,640
Speaker 4: have its servers physically close to the exchange matching engines,

457
00:28:40,240 --> 00:28:44,960
Speaker 4: the fiber optic cable that connects them is coiled so

458
00:28:45,000 --> 00:28:49,200
Speaker 4: that there's exactly the same cable length for each of

459
00:28:49,320 --> 00:28:55,440
Speaker 4: the trading firms within that data center. The other thing

460
00:28:55,440 --> 00:29:00,240
Speaker 4: that started to happen is that getting the signal from

461
00:29:00,240 --> 00:29:04,480
Speaker 4: one exchange's data center to another exchange is data center

462
00:29:05,240 --> 00:29:11,040
Speaker 4: started to become a technological speed race. Back to two

463
00:29:11,040 --> 00:29:14,720
Speaker 4: thousand and five, by and large, it would just be

464
00:29:15,520 --> 00:29:19,320
Speaker 4: sent by fiber optic cable but the exact route was

465
00:29:19,360 --> 00:29:22,920
Speaker 4: not under the control of the exchanges or of the

466
00:29:23,000 --> 00:29:25,840
Speaker 4: trading firm, so there's a lot of sort of randomness.

467
00:29:26,600 --> 00:29:30,320
Speaker 4: The crucial link here is actually between the Chicago Mercantile

468
00:29:30,400 --> 00:29:36,200
Speaker 4: Exchanges data center then in downtown's Chicago. It's now in

469
00:29:36,800 --> 00:29:43,720
Speaker 4: the suburbs of Chicago, the link between that and the

470
00:29:43,800 --> 00:29:49,520
Speaker 4: data centers that trade shares in northern New Jersey. And

471
00:29:49,880 --> 00:29:52,280
Speaker 4: there was a kind of triple evolution there. The first

472
00:29:52,280 --> 00:29:56,080
Speaker 4: evolution was that the particular trading firm and since it's

473
00:29:56,520 --> 00:29:58,880
Speaker 4: no longer directing in business, I can actually name it

474
00:29:59,000 --> 00:30:06,000
Speaker 4: get Co managed to as it were, stitched together existing

475
00:30:06,160 --> 00:30:11,680
Speaker 4: fiber optic cables to get the fastest route on the

476
00:30:11,760 --> 00:30:15,400
Speaker 4: Chicago New Jersey link, and that actually in the business

477
00:30:15,480 --> 00:30:18,440
Speaker 4: was known as the gold Line gold because of the

478
00:30:18,440 --> 00:30:21,600
Speaker 4: money that you could make by having the fastest route.

479
00:30:22,280 --> 00:30:25,600
Speaker 4: Then in twenty ten, that memory serves me right, a

480
00:30:25,720 --> 00:30:30,280
Speaker 4: new firm Spread Networks actually dug an entire new cable

481
00:30:30,760 --> 00:30:37,160
Speaker 4: from Chicago to northern New Jersey. You know, the drilling,

482
00:30:37,560 --> 00:30:40,720
Speaker 4: you know, sort of underneath car parks and you know,

483
00:30:40,880 --> 00:30:44,520
Speaker 4: just really trying to be as close to what a

484
00:30:44,520 --> 00:30:48,280
Speaker 4: geographer would call the geodesic, in other words, the fastest

485
00:30:48,360 --> 00:30:50,880
Speaker 4: route on the surface of the Earth from point eight too.

486
00:30:51,160 --> 00:30:55,720
Speaker 4: B then third phase that was trumped by the arrival

487
00:30:56,400 --> 00:30:59,840
Speaker 4: of microwave because lighting a fiber optic cable, I mean

488
00:30:59,840 --> 00:31:02,960
Speaker 4: the core of fiber optic cable is essentially a specialized

489
00:31:03,000 --> 00:31:08,440
Speaker 4: form of glass, and that glass slows the light down

490
00:31:09,000 --> 00:31:12,440
Speaker 4: to only two thirds of its speed in a vacuum,

491
00:31:12,480 --> 00:31:15,840
Speaker 4: whereas if you can shoot your electrog ginetic signal through

492
00:31:15,880 --> 00:31:19,120
Speaker 4: the atmosphere, it's not exactly at the speed of light

493
00:31:19,120 --> 00:31:21,320
Speaker 4: in the in the vacuum, but it's very very close

494
00:31:21,360 --> 00:31:23,240
Speaker 4: to the speed of light in the vacuum. So that

495
00:31:23,360 --> 00:31:27,560
Speaker 4: was the third phase. When people moved from exclusive use

496
00:31:27,560 --> 00:31:32,280
Speaker 4: of fiber optic cable to supplementing the fiber optic cable

497
00:31:32,680 --> 00:31:37,120
Speaker 4: by microwave links between Chicago and northern New Jersey.

498
00:31:37,520 --> 00:31:57,000
Speaker 3: That was incredible. I have another cultural slash market structure question,

499
00:31:57,160 --> 00:32:02,240
Speaker 3: which is, along with this intel competition to be faster

500
00:32:02,480 --> 00:32:07,600
Speaker 3: than anyone else, there was a narrative around that time

501
00:32:07,880 --> 00:32:11,600
Speaker 3: that this was a bad thing, right, And one of

502
00:32:11,680 --> 00:32:14,240
Speaker 3: the interesting cultural things is that you saw the high

503
00:32:14,240 --> 00:32:18,360
Speaker 3: frequency trading firms kind of divide themselves into good guys

504
00:32:18,400 --> 00:32:21,240
Speaker 3: and bad guys. So there were some that were saying, well,

505
00:32:21,280 --> 00:32:24,920
Speaker 3: we make liquidity right, we're good for the market. And

506
00:32:24,960 --> 00:32:28,080
Speaker 3: then there were others. I mean, they wouldn't describe themselves

507
00:32:28,120 --> 00:32:30,800
Speaker 3: this way, but they were accused of being liquidity takers

508
00:32:31,480 --> 00:32:35,800
Speaker 3: in the market. How should we think about that particular tension.

509
00:32:36,400 --> 00:32:39,600
Speaker 4: Yeah, no, that's a very fundamental thing, because you're quite right,

510
00:32:40,200 --> 00:32:46,520
Speaker 4: there's a degree of differentiation between trading firms, and indeed

511
00:32:46,560 --> 00:32:49,800
Speaker 4: in the more segregated firms, also within those firms you

512
00:32:49,800 --> 00:32:53,200
Speaker 4: have different desks fallen others on different sides of that.

513
00:32:53,280 --> 00:32:56,200
Speaker 4: I mean, the way to think about it is to

514
00:32:56,280 --> 00:33:02,080
Speaker 4: remember that in all these exchanges is organized around an

515
00:33:02,120 --> 00:33:05,360
Speaker 4: electronic order book, as I said, a list of the

516
00:33:05,400 --> 00:33:11,040
Speaker 4: bids to buy and offers to sell the instrument in question.

517
00:33:11,720 --> 00:33:17,240
Speaker 4: What a market making firm does is it places lots

518
00:33:17,320 --> 00:33:22,600
Speaker 4: of orders into the order book at prices that can't

519
00:33:22,600 --> 00:33:29,200
Speaker 4: immediately be executed. But those orders then populate the order book.

520
00:33:29,760 --> 00:33:33,760
Speaker 4: So if somebody else comes along, an individual investor or

521
00:33:33,880 --> 00:33:38,000
Speaker 4: maybe an asset management firm, somebody else comes along and

522
00:33:38,160 --> 00:33:42,640
Speaker 4: needs to trade, they find an order book that's populated

523
00:33:42,720 --> 00:33:47,760
Speaker 4: with lots of existing bids and offers that they can

524
00:33:47,960 --> 00:33:53,600
Speaker 4: execute against. So the market making firm is providing liquidity,

525
00:33:54,640 --> 00:33:59,320
Speaker 4: and most people reckon that that's a good thing. The

526
00:33:59,400 --> 00:34:03,400
Speaker 4: other kind of firm, the one whereas you say, and

527
00:34:03,440 --> 00:34:07,840
Speaker 4: there tends to be more controversy, they don't do that.

528
00:34:08,000 --> 00:34:12,080
Speaker 4: They don't constantly populate the order book. Their systems constantly

529
00:34:12,280 --> 00:34:17,799
Speaker 4: monitor the order book, and then when they detect what

530
00:34:18,040 --> 00:34:24,040
Speaker 4: they think is a profitable opportunity, they execute against the

531
00:34:24,200 --> 00:34:28,120
Speaker 4: orders that are already in the order book, and that

532
00:34:28,400 --> 00:34:32,200
Speaker 4: is called liquidity taking, because of course the execution removes

533
00:34:32,239 --> 00:34:37,360
Speaker 4: the order from the order book. And then in practice,

534
00:34:37,400 --> 00:34:40,919
Speaker 4: a lot of this is actually going on between high

535
00:34:40,960 --> 00:34:43,319
Speaker 4: flugency trading firms, because most of the orders in the

536
00:34:43,400 --> 00:34:47,800
Speaker 4: order book are placed by market making high flugency trading firms,

537
00:34:48,200 --> 00:34:52,120
Speaker 4: and a lot of the executions against those are by

538
00:34:52,960 --> 00:34:59,600
Speaker 4: the high flugency trading firms that specialize in taking liquidity.

539
00:35:00,560 --> 00:35:05,200
Speaker 4: And this is the core of what gives the business

540
00:35:05,600 --> 00:35:10,640
Speaker 4: its characteristic as an arms race in speed. So for

541
00:35:10,719 --> 00:35:15,440
Speaker 4: imagine for a moment that your algorithms are trading equities

542
00:35:15,480 --> 00:35:18,960
Speaker 4: in one of the data centers in northern New Jersey,

543
00:35:19,680 --> 00:35:28,560
Speaker 4: and the relevant stock index future traded in Chicago changes

544
00:35:28,680 --> 00:35:30,800
Speaker 4: in price or even you know, there's a big shift

545
00:35:30,840 --> 00:35:34,000
Speaker 4: in the order book for that stock index future, and

546
00:35:34,040 --> 00:35:37,799
Speaker 4: then let's say the price of the stock index future falls,

547
00:35:38,840 --> 00:35:44,120
Speaker 4: that's very likely to lead within a tiny fraction of

548
00:35:44,120 --> 00:35:48,600
Speaker 4: a second, to falls in the price of the underlying

549
00:35:48,800 --> 00:35:53,800
Speaker 4: shares being traded in New Jersey. And in that tiny

550
00:35:53,880 --> 00:35:57,440
Speaker 4: little fraction of a second, in that intervening period, a

551
00:35:57,440 --> 00:36:00,919
Speaker 4: lot of the market making firms or in those order

552
00:36:00,960 --> 00:36:05,880
Speaker 4: books become stale as people in the markets. So the

553
00:36:05,960 --> 00:36:12,680
Speaker 4: making firms rush to try to cancel their stale orders,

554
00:36:13,080 --> 00:36:20,760
Speaker 4: and the taking firms race to execute against those stale orders.

555
00:36:21,120 --> 00:36:23,560
Speaker 4: And that's a race that nowadays can literally be played

556
00:36:23,600 --> 00:36:24,760
Speaker 4: out in nanoseconds.

557
00:36:25,200 --> 00:36:28,799
Speaker 2: That's interesting. I hadn't appreciated that dynamic at all, to

558
00:36:28,840 --> 00:36:33,000
Speaker 2: be honest. You mentioned that these high frequency trading firms

559
00:36:33,239 --> 00:36:36,240
Speaker 2: they're all sort of like they're run and operated basically

560
00:36:36,280 --> 00:36:39,359
Speaker 2: by the employees, the owner or something like that. What

561
00:36:39,520 --> 00:36:42,520
Speaker 2: is it that sort of distinguishes a high frequency trading

562
00:36:42,520 --> 00:36:46,040
Speaker 2: firm from say a hedge fund that would have limited

563
00:36:46,080 --> 00:36:49,399
Speaker 2: partners and make distributions, et cetera. You know, you never

564
00:36:49,520 --> 00:36:51,800
Speaker 2: hear about like, oh, I have an investment with Jane

565
00:36:51,800 --> 00:36:54,200
Speaker 2: Street or something like that. I don't think that's right.

566
00:36:54,280 --> 00:36:56,200
Speaker 2: I think what is it about the nature of the

567
00:36:56,239 --> 00:36:59,480
Speaker 2: business such that essentially they trade their own capital.

568
00:36:59,760 --> 00:37:02,280
Speaker 4: Yeah, I guess that's just you know, in a sense,

569
00:37:02,640 --> 00:37:05,359
Speaker 4: it's one of those things that's historical evolution. I mean,

570
00:37:05,440 --> 00:37:10,960
Speaker 4: in many cases, those hyperingsy trading firms were initially set

571
00:37:11,040 --> 00:37:17,600
Speaker 4: up by successful floor traders, particularly floor traders in the

572
00:37:17,680 --> 00:37:23,520
Speaker 4: Chicago markets, the futures markets, and if you were successful

573
00:37:23,520 --> 00:37:27,680
Speaker 4: in that, you could you might not become a billionaire,

574
00:37:27,760 --> 00:37:31,440
Speaker 4: but you know, you could make decent amount of money,

575
00:37:31,800 --> 00:37:35,840
Speaker 4: tens of millions of dollars, and that was enough to

576
00:37:36,000 --> 00:37:42,160
Speaker 4: enable you to start an initially small automated trading firm.

577
00:37:42,200 --> 00:37:45,680
Speaker 4: And you often didn't need any external capital to do it.

578
00:37:45,719 --> 00:37:50,520
Speaker 4: You just had to buy the necessary technical kit and

579
00:37:50,760 --> 00:37:54,560
Speaker 4: hire technically savvy people and so on. But you could

580
00:37:54,560 --> 00:37:57,359
Speaker 4: start really quite small. You could start, you know, ten

581
00:37:57,400 --> 00:38:00,719
Speaker 4: person firm or something like that. Now the business has

582
00:38:00,719 --> 00:38:03,560
Speaker 4: got a lot more expensive since then because in good

583
00:38:03,560 --> 00:38:06,960
Speaker 4: part of the speed race that we've just been talking about.

584
00:38:07,000 --> 00:38:13,040
Speaker 4: But by and large those firms made profits. The owners

585
00:38:13,120 --> 00:38:16,480
Speaker 4: reinvested the profits in the firms, so that you know,

586
00:38:16,520 --> 00:38:21,640
Speaker 4: the capabilities of the firms grew to meet the growing

587
00:38:22,280 --> 00:38:25,960
Speaker 4: demands on them. And then another thing that should perhaps

588
00:38:26,040 --> 00:38:29,960
Speaker 4: be said is that those founders, which have the majority

589
00:38:30,040 --> 00:38:35,799
Speaker 4: of their net worth invested in the firm and so

590
00:38:36,719 --> 00:38:43,439
Speaker 4: their risk control was often pretty good. You know, risk

591
00:38:43,520 --> 00:38:45,839
Speaker 4: management for those firms was not a sort of you know,

592
00:38:46,120 --> 00:38:49,799
Speaker 4: a separate bureaucratic department that the traders had to try

593
00:38:49,840 --> 00:38:53,240
Speaker 4: to outwit, so to speak. The founder would quickly detect

594
00:38:53,320 --> 00:38:57,359
Speaker 4: if you were trying to do something like that. Now,

595
00:38:57,840 --> 00:39:01,799
Speaker 4: some automated trading firms have blown up, nevertheless, but it's

596
00:39:01,800 --> 00:39:06,200
Speaker 4: actually quite interesting how few of them have blown up, Yeah,

597
00:39:06,200 --> 00:39:10,880
Speaker 4: because of you know, for example, because of classical software bugs.

598
00:39:11,480 --> 00:39:16,080
Speaker 4: And that's I think because the relatively small structure, the

599
00:39:16,440 --> 00:39:21,880
Speaker 4: hands on involvement of the founder's et cetera, has created

600
00:39:22,680 --> 00:39:30,399
Speaker 4: a kind technical culture that actually works pretty reasonably well

601
00:39:30,600 --> 00:39:32,560
Speaker 4: better than I would have expected it to work at

602
00:39:32,600 --> 00:39:33,640
Speaker 4: the start of this research.

603
00:39:35,320 --> 00:39:38,160
Speaker 3: So one thing I wanted to ask is this idea

604
00:39:38,200 --> 00:39:41,120
Speaker 3: of physical limitations to how fast we can actually go.

605
00:39:41,880 --> 00:39:44,000
Speaker 3: I'm pretty sure people always say that you can't go

606
00:39:44,120 --> 00:39:47,560
Speaker 3: faster than the speed of light. There's probably some caveats there,

607
00:39:47,600 --> 00:39:51,120
Speaker 3: about like quantum entanglement and stuff like that. But surely

608
00:39:51,840 --> 00:39:56,439
Speaker 3: we must be getting close to how fast things can

609
00:39:56,520 --> 00:39:59,520
Speaker 3: actually go. What's your sense of how long the speed

610
00:39:59,600 --> 00:40:00,719
Speaker 3: race can continue?

611
00:40:01,120 --> 00:40:04,920
Speaker 4: Yeah, well, we can never get to zero, of course,

612
00:40:05,280 --> 00:40:09,680
Speaker 4: I think Einstein is basically correct. You can't get faster

613
00:40:10,040 --> 00:40:15,359
Speaker 4: than the speed of light in a vacuum, and similarly

614
00:40:15,920 --> 00:40:19,839
Speaker 4: a computer system, there's always going to be a non

615
00:40:20,000 --> 00:40:26,640
Speaker 4: zero processing time of the system, but you can get

616
00:40:27,400 --> 00:40:32,479
Speaker 4: ever ever closer to that. So in mathematics speak, zero

617
00:40:32,680 --> 00:40:35,799
Speaker 4: is an sym topic in it. That's to say, you

618
00:40:35,840 --> 00:40:38,880
Speaker 4: can always get closer to it, but you're never actually

619
00:40:38,920 --> 00:40:41,840
Speaker 4: going to get right there. And I think that's the

620
00:40:41,920 --> 00:40:45,719
Speaker 4: nature of the business. You know, we're still in the

621
00:40:45,800 --> 00:40:51,719
Speaker 4: nano second regime. I remember correctly, the next lowest time

622
00:40:51,760 --> 00:40:55,960
Speaker 4: interval is the pico second. You know, I could imagine

623
00:40:55,960 --> 00:41:00,480
Speaker 4: this continuing in a domain of pico seconds. So that's

624
00:41:00,520 --> 00:41:02,879
Speaker 4: the way I would see it, that there's a hard

625
00:41:02,920 --> 00:41:04,960
Speaker 4: limit that we're never actually going to get to the

626
00:41:05,000 --> 00:41:07,759
Speaker 4: hard limit. We're still going to race to get as

627
00:41:07,800 --> 00:41:08,719
Speaker 4: close as possible.

628
00:41:09,280 --> 00:41:12,560
Speaker 2: But you're understanding, as the time of your work is

629
00:41:12,600 --> 00:41:15,400
Speaker 2: that the race is not over that for these firm

630
00:41:15,800 --> 00:41:18,640
Speaker 2: and I'm sure they have many different projects that they're played,

631
00:41:18,719 --> 00:41:21,239
Speaker 2: including various things with AI which we haven't gotten to

632
00:41:21,320 --> 00:41:23,560
Speaker 2: and I guess we won't. But the speed race is

633
00:41:23,640 --> 00:41:25,359
Speaker 2: not and will never be done.

634
00:41:25,640 --> 00:41:29,080
Speaker 4: Yeah, I think that's correct. Now. Of course, there's an

635
00:41:29,120 --> 00:41:34,080
Speaker 4: economic process, yeah at work here, which is to say

636
00:41:34,440 --> 00:41:41,840
Speaker 4: that the investments that you make in speed have to

637
00:41:41,880 --> 00:41:49,040
Speaker 4: be recoupable from the trading profits that you make from

638
00:41:49,120 --> 00:41:52,120
Speaker 4: your trading. And I think Tracy I think said at

639
00:41:52,120 --> 00:41:56,040
Speaker 4: the very beginning what essentially is going on here is

640
00:41:56,080 --> 00:42:00,920
Speaker 4: that structural features of financial markets are being exploited, like

641
00:42:01,320 --> 00:42:07,000
Speaker 4: the relationship between the stock index future and the underlying equities.

642
00:42:07,560 --> 00:42:12,320
Speaker 4: The amount of money to be made by exploiting those

643
00:42:12,800 --> 00:42:18,160
Speaker 4: kind of structural features is not trivial. It's been nicely

644
00:42:18,200 --> 00:42:24,120
Speaker 4: measured by the Chicago economist Eric Suffert and colleagues. It's

645
00:42:24,120 --> 00:42:29,759
Speaker 4: not trivial, but it's perhaps single digit billions of dollars.

646
00:42:30,360 --> 00:42:35,280
Speaker 4: So you know, suddenly deciding you're going to invest fifty

647
00:42:35,320 --> 00:42:38,960
Speaker 4: billion dollars and the technology of speed would be a

648
00:42:39,040 --> 00:42:42,440
Speaker 4: dumb thing to do because you wouldn't be able to

649
00:42:42,600 --> 00:42:47,680
Speaker 4: recupe it. So there is that economic factor that is

650
00:42:48,040 --> 00:42:53,600
Speaker 4: I'm pretty certain slowing. The speed race is still there,

651
00:42:54,040 --> 00:42:57,160
Speaker 4: but it's it's you know, and things are getting faster,

652
00:42:57,280 --> 00:42:59,840
Speaker 4: but the rate at which they're getting faster, yeah, certainly

653
00:42:59,840 --> 00:43:03,799
Speaker 4: not accelerating. And I think that economic factor probably explains it.

654
00:43:03,960 --> 00:43:04,680
Speaker 2: That makes sense.

655
00:43:04,960 --> 00:43:07,680
Speaker 3: So we should talk about the impact of HFT on

656
00:43:07,920 --> 00:43:10,279
Speaker 3: the overall market a little bit more. And one of

657
00:43:10,280 --> 00:43:13,160
Speaker 3: the things that caught my eye in your book was

658
00:43:13,360 --> 00:43:16,120
Speaker 3: you cite a previous study I can't remember by who,

659
00:43:16,520 --> 00:43:20,240
Speaker 3: but basically saying that the efficiency of financial markets has

660
00:43:20,239 --> 00:43:25,440
Speaker 3: not improved between the eighteen eighties and twenty twelve, which

661
00:43:25,719 --> 00:43:28,560
Speaker 3: is very countering a graduate.

662
00:43:28,640 --> 00:43:29,480
Speaker 2: But what does that mean?

663
00:43:29,640 --> 00:43:33,960
Speaker 4: Yeah, so that does work by Thomas Philippont or his

664
00:43:34,120 --> 00:43:38,560
Speaker 4: friend shop pronounces in the American ways Thomas Philippon what

665
00:43:38,640 --> 00:43:41,440
Speaker 4: he means by efficiency there is really rather different from

666
00:43:41,680 --> 00:43:44,359
Speaker 4: what I've been talking about. What he means by efficiency

667
00:43:44,719 --> 00:43:48,960
Speaker 4: is what he calls the unit cost of financial intermediation,

668
00:43:49,880 --> 00:43:54,799
Speaker 4: which essentially is, you know, basically putting it crudely, how

669
00:43:54,880 --> 00:43:59,040
Speaker 4: much it costs to do the kind of thing that

670
00:44:00,080 --> 00:44:04,600
Speaker 4: investors want to do, the asset managers want to do,

671
00:44:05,360 --> 00:44:09,800
Speaker 4: and so on. And you know, it's very striking finding

672
00:44:10,239 --> 00:44:13,480
Speaker 4: that from the eighteen eighties to I think his most

673
00:44:13,520 --> 00:44:17,640
Speaker 4: recent data goes up to twenty fifteen, that there was

674
00:44:17,840 --> 00:44:22,239
Speaker 4: no really clear cut tendency for that cost to decline

675
00:44:22,719 --> 00:44:28,799
Speaker 4: despite all the advances and information and communication technologies over

676
00:44:28,840 --> 00:44:34,320
Speaker 4: those many decades. And the explanation, to put it simplistically

677
00:44:34,719 --> 00:44:43,040
Speaker 4: and crudely, is the capture of those efficiency gains in

678
00:44:43,160 --> 00:44:49,319
Speaker 4: the form of high pay in the financial sector, you know,

679
00:44:49,360 --> 00:44:53,440
Speaker 4: typically through the form of fees, so the fees that

680
00:44:53,480 --> 00:44:56,480
Speaker 4: you pay for an index fancy for example, those have

681
00:44:56,560 --> 00:45:00,120
Speaker 4: really gone down. But people have also been moving their

682
00:45:00,360 --> 00:45:03,600
Speaker 4: money into private equity and the like hedge fans which

683
00:45:03,600 --> 00:45:06,279
Speaker 4: have much higher fees and so on, so those kind

684
00:45:06,320 --> 00:45:11,120
Speaker 4: of effects seem to have sort of canceled themselves out.

685
00:45:11,320 --> 00:45:17,040
Speaker 4: In the nineteen forties, a professional in finance was basically

686
00:45:17,080 --> 00:45:22,040
Speaker 4: paid roughly the same as somebody with equivalent educational qualifications

687
00:45:22,080 --> 00:45:25,520
Speaker 4: in a different line of business, and then from the

688
00:45:25,600 --> 00:45:32,080
Speaker 4: nineteen seventies onwards the gap has got bigger and bigger

689
00:45:32,880 --> 00:45:36,839
Speaker 4: and bigger. Now these days, of course, you can make

690
00:45:36,840 --> 00:45:40,480
Speaker 4: a lot of money being a technologist in AI for example,

691
00:45:40,520 --> 00:45:45,560
Speaker 4: but by and large those sort of exceptions aside. Finance

692
00:45:45,680 --> 00:45:49,880
Speaker 4: is an extraordinarily well paid professional at least for those

693
00:45:49,920 --> 00:45:53,000
Speaker 4: in the central roles in it. And that's essentially the

694
00:45:53,080 --> 00:45:55,840
Speaker 4: explanation of that finding by Philippon.

695
00:45:56,320 --> 00:45:58,680
Speaker 3: Well, this is a good opportunity to ask about AI,

696
00:45:58,880 --> 00:46:02,960
Speaker 3: because I suppose it's inevitable. As a sociologist who examines

697
00:46:03,360 --> 00:46:06,960
Speaker 3: the tech industry or looks at how tech is impacting things,

698
00:46:07,520 --> 00:46:09,080
Speaker 3: your next project must be AI.

699
00:46:09,360 --> 00:46:10,839
Speaker 4: Right, Yes it is.

700
00:46:11,280 --> 00:46:14,279
Speaker 3: And what's the particular angle or what have you been

701
00:46:14,320 --> 00:46:15,359
Speaker 3: discovering so far?

702
00:46:15,880 --> 00:46:18,560
Speaker 4: Yeah, it's very early days. But the thing I'm most

703
00:46:18,600 --> 00:46:25,640
Speaker 4: interested in so far is the question of scaling a I.

704
00:46:25,920 --> 00:46:28,880
Speaker 4: Because of course it's no secret to anybody who reads

705
00:46:29,120 --> 00:46:34,399
Speaker 4: reads the newspaper, subscribes to Bloomberg or whatever. I mean,

706
00:46:34,440 --> 00:46:40,240
Speaker 4: the huge trillions of dollars are being thrown at AI infrastructure.

707
00:46:40,760 --> 00:46:44,320
Speaker 4: And absolutely there is a sort of logic there that's

708
00:46:44,400 --> 00:46:50,560
Speaker 4: repeatedly stated that these systems are all built around neural networks,

709
00:46:51,040 --> 00:46:55,680
Speaker 4: and the effectiveness of a neural network grows with the

710
00:46:55,840 --> 00:47:00,520
Speaker 4: size of the network, the size of the trade data,

711
00:47:00,640 --> 00:47:03,799
Speaker 4: the number of parameters in the model, and so on.

712
00:47:03,880 --> 00:47:09,480
Speaker 4: Then there are well known scaling laws. But this is

713
00:47:09,600 --> 00:47:13,200
Speaker 4: the thing that interests me is the But it's a

714
00:47:13,239 --> 00:47:16,920
Speaker 4: very nice little statement from Sam Altman in February of

715
00:47:17,680 --> 00:47:24,400
Speaker 4: last year that the intelligence of a system is roughly

716
00:47:25,120 --> 00:47:30,120
Speaker 4: the log the logarithm, in other words, of the resources

717
00:47:30,680 --> 00:47:35,960
Speaker 4: devoted to training it, running it, to computation that inferenced time,

718
00:47:36,600 --> 00:47:40,320
Speaker 4: and so on. Now, of course, what Altman meant was basically,

719
00:47:41,200 --> 00:47:44,680
Speaker 4: give the industry more money and you'll get more intelligence.

720
00:47:45,239 --> 00:47:48,000
Speaker 4: And that's, of course, indeed, the giving more money to

721
00:47:48,040 --> 00:47:52,440
Speaker 4: the industry is exactly what's going on. But a logarithmic function,

722
00:47:53,040 --> 00:47:56,200
Speaker 4: there's a bit of mass here. A logarithmic function, at

723
00:47:56,320 --> 00:48:00,520
Speaker 4: least of the kind that Altman is referring to, is

724
00:48:00,560 --> 00:48:06,160
Speaker 4: a diminishing returns function. You can draw its graph and

725
00:48:06,640 --> 00:48:11,480
Speaker 4: it very clearly demonstrates diminishing returns. You can always get

726
00:48:11,480 --> 00:48:16,960
Speaker 4: better and better, but each increment costs you more in

727
00:48:17,120 --> 00:48:23,040
Speaker 4: terms of the resources deployed. And we're dealing here where

728
00:48:23,320 --> 00:48:25,719
Speaker 4: the horizontal axis in the graph, so to speak, is

729
00:48:25,760 --> 00:48:34,520
Speaker 4: denominated in trillions of dollars of financial input or hundreds

730
00:48:34,560 --> 00:48:39,719
Speaker 4: of megatons of carbon dioxide emitted by the electricity generation

731
00:48:39,960 --> 00:48:45,760
Speaker 4: needed to power the data center. So the question becomes,

732
00:48:46,239 --> 00:48:50,839
Speaker 4: on a diminishing returns curve, how far do you go

733
00:48:51,960 --> 00:48:56,520
Speaker 4: when do you decide we really got to stop? Can

734
00:48:56,560 --> 00:49:01,160
Speaker 4: you decide we really got to stop? Do you have

735
00:49:01,360 --> 00:49:05,160
Speaker 4: kind of quasi magical beliefs so to speak, that at

736
00:49:05,200 --> 00:49:09,880
Speaker 4: some point the diminishing returns, something qualitative will happen, that

737
00:49:10,400 --> 00:49:16,040
Speaker 4: artificial general intelligence or super intelligence will suddenly appear. So

738
00:49:16,120 --> 00:49:19,080
Speaker 4: that's the core of what I'm interested in right now.

739
00:49:19,719 --> 00:49:23,760
Speaker 4: How far do you go along a diminishing returns curve?

740
00:49:24,239 --> 00:49:26,080
Speaker 3: Joe, I just want to state for the record, if

741
00:49:26,080 --> 00:49:28,319
Speaker 3: you give me more money, I get more intelligent. There's

742
00:49:28,320 --> 00:49:30,200
Speaker 3: no diminishing returns, just.

743
00:49:30,160 --> 00:49:35,360
Speaker 2: For the record, you know, noted first of all. And

744
00:49:35,440 --> 00:49:37,840
Speaker 2: it's interesting, you know, hearing this in the context, and

745
00:49:37,880 --> 00:49:39,920
Speaker 2: it suddenly makes so much sense. How this visit to

746
00:49:39,960 --> 00:49:42,799
Speaker 2: your work and this idea of like the arms race, right,

747
00:49:42,960 --> 00:49:46,000
Speaker 2: because yes, it's true, like maybe there's only so much

748
00:49:46,320 --> 00:49:50,360
Speaker 2: extra profit available for the firm, and maybe that pool

749
00:49:50,400 --> 00:49:53,360
Speaker 2: of profit is shrinking, and maybe it gets more and

750
00:49:53,400 --> 00:49:57,800
Speaker 2: more costly to sort of exploit the remaining profit that's available.

751
00:49:57,800 --> 00:50:00,759
Speaker 2: But on the other hand, you can't fall off, you

752
00:50:00,800 --> 00:50:03,760
Speaker 2: can't let and this is so there. It's true in HFT,

753
00:50:03,920 --> 00:50:07,040
Speaker 2: and it's clearly true in AI, where Okay, like it

754
00:50:07,120 --> 00:50:10,239
Speaker 2: spends more and more money to improve the model, but

755
00:50:10,360 --> 00:50:13,480
Speaker 2: you can't fall behind, even if the economics look worse

756
00:50:13,520 --> 00:50:18,919
Speaker 2: with each iteration. Anyway, Professor McKenzie, lovely conversation. I really

757
00:50:19,040 --> 00:50:21,399
Speaker 2: enjoyed that. We really learned a lot. Really appreciate you

758
00:50:21,480 --> 00:50:23,960
Speaker 2: coming on to odd loads and a yeah, thank.

759
00:50:23,800 --> 00:50:25,880
Speaker 4: You, well, thank you both, thank you both for inviting me,

760
00:50:26,000 --> 00:50:28,839
Speaker 4: like I said, and you know, thanks for a really

761
00:50:28,920 --> 00:50:29,680
Speaker 4: great conversation.

762
00:50:30,000 --> 00:50:30,719
Speaker 2: It's fantastic.

763
00:50:30,840 --> 00:50:32,600
Speaker 3: I have to have you back on when you publish

764
00:50:32,640 --> 00:50:33,359
Speaker 3: your AI book.

765
00:50:33,480 --> 00:50:51,040
Speaker 2: Absolutely racing. I love that conversation really interesting. I love

766
00:50:51,160 --> 00:50:55,439
Speaker 2: like encountering people who are like actually understand the tech,

767
00:50:55,600 --> 00:51:00,440
Speaker 2: actually can articulate what the tech is doing. Actually, it's

768
00:51:00,440 --> 00:51:04,239
Speaker 2: always impressive someone with the sociology background to just sort

769
00:51:04,239 --> 00:51:06,640
Speaker 2: of be like that comfortable, and I think that's like

770
00:51:06,680 --> 00:51:08,759
Speaker 2: the through line of his work is like he gets

771
00:51:08,840 --> 00:51:09,640
Speaker 2: it right.

772
00:51:09,800 --> 00:51:13,120
Speaker 3: So in the book, there's lots of like field trips

773
00:51:13,200 --> 00:51:16,000
Speaker 3: to data centers and looking at cables and things like that,

774
00:51:16,040 --> 00:51:18,720
Speaker 3: but then also, as he stated, just talking to people

775
00:51:18,920 --> 00:51:22,279
Speaker 3: and getting anecdotes and there's funny stories about like the

776
00:51:22,320 --> 00:51:24,799
Speaker 3: Battle of the Asterisks and things like that. That's at

777
00:51:24,840 --> 00:51:26,879
Speaker 3: the very end, but people should go and read it.

778
00:51:27,400 --> 00:51:29,000
Speaker 3: The other thing that stood out to me from that

779
00:51:29,040 --> 00:51:33,239
Speaker 3: conversation was towards the end when we discussed AI, you

780
00:51:33,320 --> 00:51:36,400
Speaker 3: made the point that you get this similar dynamic between

781
00:51:36,440 --> 00:51:39,480
Speaker 3: the HFT and AI now where because everything is framed

782
00:51:39,960 --> 00:51:44,200
Speaker 3: as existential, you just can't stop, right, You always have

783
00:51:44,239 --> 00:51:45,360
Speaker 3: to keep going totally.

784
00:51:45,360 --> 00:51:48,720
Speaker 2: Hey, look, I mean I suppose the high frequency trading

785
00:51:48,840 --> 00:51:52,560
Speaker 2: is not sort of like existential in the broad sense,

786
00:51:52,600 --> 00:51:57,439
Speaker 2: but it's the level, yeah, at the firm level. So

787
00:51:57,480 --> 00:51:59,239
Speaker 2: it's like, and I hadn't really thought about it. Okay,

788
00:51:59,280 --> 00:52:02,560
Speaker 2: you like have this like pool of theoretical profit, which

789
00:52:02,640 --> 00:52:05,799
Speaker 2: is the gap between where the futures are trading at

790
00:52:05,840 --> 00:52:09,479
Speaker 2: Chicago and where these stocks are trading in New York.

791
00:52:09,840 --> 00:52:13,480
Speaker 2: That's fixed, right, that's not gonna get that big. But again,

792
00:52:14,080 --> 00:52:17,480
Speaker 2: like someone gets faster at exploiting that. And I had

793
00:52:17,520 --> 00:52:21,040
Speaker 2: never really heard quite until your question and his answer,

794
00:52:21,600 --> 00:52:24,719
Speaker 2: this sort of maker taker dynamic of Okay, I have

795
00:52:24,800 --> 00:52:27,560
Speaker 2: these orders and now I'm quickly rushing to cancel them,

796
00:52:27,600 --> 00:52:31,560
Speaker 2: and you're quickly rushing to fulfill them. And if you

797
00:52:31,600 --> 00:52:34,319
Speaker 2: and I are both in the market, we can't slow

798
00:52:34,360 --> 00:52:36,799
Speaker 2: If you get faster I must get faster because you're

799
00:52:36,800 --> 00:52:39,600
Speaker 2: gonna snipe me every time, or vice versa, et cetera.

800
00:52:40,120 --> 00:52:42,120
Speaker 2: But you know, they still make a lot of money.

801
00:52:42,160 --> 00:52:45,239
Speaker 2: It seems like, unlike the AI firms, they make at

802
00:52:45,239 --> 00:52:46,560
Speaker 2: the moment. Yeah, exactly.

803
00:52:46,640 --> 00:52:49,520
Speaker 3: That was a funny dynamic in HFT world, the like

804
00:52:49,640 --> 00:52:53,200
Speaker 3: accusations of taker versus maker, and I always think of

805
00:52:53,239 --> 00:52:55,480
Speaker 3: the you know, the Spider Man meme where they're all

806
00:52:55,560 --> 00:52:58,040
Speaker 3: kind of rowing each other. Yeah, it felt very much

807
00:52:58,080 --> 00:53:00,239
Speaker 3: like that. But it was really great to catch on

808
00:53:00,400 --> 00:53:03,279
Speaker 3: HFT again. This yeah blast from the past because you

809
00:53:03,360 --> 00:53:05,200
Speaker 3: used to hear about it more and now it's become

810
00:53:05,280 --> 00:53:08,080
Speaker 3: so normalized that people just don't talk about it that much.

811
00:53:08,200 --> 00:53:10,279
Speaker 2: Well, you know, we heard about it a lot, and

812
00:53:10,520 --> 00:53:13,520
Speaker 2: especially post two thousand and eight. That was the ultimate

813
00:53:13,680 --> 00:53:16,759
Speaker 2: finger pointing era, right Michael Lewis, because you just have

814
00:53:16,920 --> 00:53:20,240
Speaker 2: like everyone had some Oh it's the naked short sellers,

815
00:53:20,320 --> 00:53:23,920
Speaker 2: it's the credit rating agency, it's the uh, you know,

816
00:53:24,160 --> 00:53:27,919
Speaker 2: the law that forces banks to be equitable and who

817
00:53:27,920 --> 00:53:30,680
Speaker 2: they distribute mortgages to, et cetera. Like there was a

818
00:53:30,719 --> 00:53:34,000
Speaker 2: million finger pointing. Oh, maybe it's the HFT firms, Maybe

819
00:53:34,000 --> 00:53:36,719
Speaker 2: it's the short sellars. Whatever. So, I mean, part of

820
00:53:36,760 --> 00:53:38,320
Speaker 2: the reason we don't hear about it as much is

821
00:53:38,360 --> 00:53:40,959
Speaker 2: because there hasn't been a crisis, et cetera.

822
00:53:41,040 --> 00:53:43,080
Speaker 4: But as you said, the race continues.

823
00:53:43,239 --> 00:53:44,719
Speaker 2: They're of various flavors.

824
00:53:44,880 --> 00:53:46,600
Speaker 3: You can never get to zero, but you can always

825
00:53:46,640 --> 00:53:47,200
Speaker 3: get closer.

826
00:53:47,760 --> 00:53:47,920
Speaker 4: You know.

827
00:53:48,000 --> 00:53:50,600
Speaker 2: It's like, you know, I always think about some lines.

828
00:53:50,640 --> 00:53:53,680
Speaker 2: They go like straight up, you know, it's like they

829
00:53:53,719 --> 00:53:56,120
Speaker 2: ever like curved back around, et cetera. You get like

830
00:53:56,239 --> 00:53:59,000
Speaker 2: negative space, like can we do even better than the

831
00:53:59,000 --> 00:54:02,400
Speaker 2: line go up? Like line go up and backwards? I

832
00:54:02,400 --> 00:54:05,319
Speaker 2: guess Einstein would say, Now, if only we.

833
00:54:05,280 --> 00:54:08,040
Speaker 3: Could have Einstein on a guest to talk about trading.

834
00:54:08,120 --> 00:54:09,720
Speaker 2: Oh, talk about Hi, you forget trading?

835
00:54:10,000 --> 00:54:10,799
Speaker 3: Shall we leave it there?

836
00:54:10,840 --> 00:54:12,279
Speaker 2: That would be a perfect guest. Let's leave it there,

837
00:54:12,280 --> 00:54:12,719
Speaker 2: all right.

838
00:54:12,880 --> 00:54:15,360
Speaker 3: This has been another episode of the oud Loots podcast.

839
00:54:15,480 --> 00:54:18,680
Speaker 3: I'm Tracy Alloway. You can follow me at Tracy Alloway.

840
00:54:18,440 --> 00:54:21,400
Speaker 2: And I'm Joe Wisenthal. You can follow me at the Stalwart.

841
00:54:21,600 --> 00:54:24,960
Speaker 2: Check out Donald McKenzie's book Trading at the Speed of Light,

842
00:54:25,040 --> 00:54:28,280
Speaker 2: and of course follow our producers Carmen Rodriguez at Carman

843
00:54:28,400 --> 00:54:31,719
Speaker 2: Ermann Dashill, Bennett at Dashbot and Kill Brooks at Kilbrooks

844
00:54:31,800 --> 00:54:34,040
Speaker 2: and from our Odd Lots content. Go to Bloomberg dot

845
00:54:34,080 --> 00:54:37,160
Speaker 2: com slash odd Lots with the daily newsletter and all

846
00:54:37,200 --> 00:54:39,560
Speaker 2: of our episodes, and you can chat about all these

847
00:54:39,600 --> 00:54:42,839
Speaker 2: topics twenty four to seven in our discord Discord dot

848
00:54:42,880 --> 00:54:44,200
Speaker 2: gg slash odd Lots.

849
00:54:44,480 --> 00:54:46,640
Speaker 3: And if you enjoy odd Lots, if you like it

850
00:54:46,680 --> 00:54:49,560
Speaker 3: when we look back at the HFT boom and how

851
00:54:49,560 --> 00:54:52,280
Speaker 3: it continues, I guess, then please leave us a positive

852
00:54:52,320 --> 00:54:55,400
Speaker 3: review on your favorite podcast platform. And remember, if you

853
00:54:55,400 --> 00:54:57,719
Speaker 3: are a Bloomberg subscriber, you can listen to all of

854
00:54:57,719 --> 00:55:00,759
Speaker 3: our episodes absolutely add free. All you need to do

855
00:55:00,880 --> 00:55:03,680
Speaker 3: is find the Bloomberg channel on Apple Podcasts and follow

856
00:55:03,719 --> 00:55:06,000
Speaker 3: the instructions there. Thanks for listening