النص 🔗
WEBVTT
NOTE
Transcription provided by Podhome.fm
Created: 7/6/2024 5:41:36 PM
Duration: 4230.408
Channels: 1
1
00:00:13.885 --> 00:00:15.505
Hello, and welcome to podcast.init,
2
00:00:16.125 --> 00:00:25.160
the podcast about Python and the people who make it great. You can subscribe to our show on iTunes, Stitcher, TuneIn Radio, or add our RSS feed to your podcatcher of choice.
3
00:00:25.619 --> 00:00:29.144
You can also follow us on Twitter or Google Plus, and please give us feedback.
4
00:00:29.605 --> 00:00:41.690
You can leave a review on Itunes to help other people find the show, send us a tweet, send us an email, leave us a message on Google plus, you can leave a comment on our show notes, or you can also visit our new discourse forum at discourse.pythonpodcast.com.
5
00:00:44.550 --> 00:00:52.575
I would like to thank everyone who has donated to the show. Your contributions help us make the show sustainable. For details on how to support the show, you can visit our site at pythonpodcast.com.
6
00:00:55.010 --> 00:00:59.989
I would also like to thank Hired, a job marketplace for developers, for sponsoring this episode of podcast.init.
7
00:01:00.930 --> 00:01:02.645
Use the link hired. Com/podcastinit
8
00:01:03.425 --> 00:01:06.005
to double your signing bonus and get $4, 000.
9
00:01:06.945 --> 00:01:12.490
We are recording today on December 30th 20 15, and your host as usual are Tobias Macy and Chris Patti.
10
00:01:12.790 --> 00:01:18.490
Today, we are interviewing Yves Hilpisch about quantitative finance. Yves, could you please introduce yourself?
11
00:01:19.495 --> 00:01:23.275
Well, first of all, thank you for having me. Yeah, my name is Yves.
12
00:01:23.575 --> 00:01:24.475
I said it already.
13
00:01:24.935 --> 00:01:25.675
I'm German.
14
00:01:26.775 --> 00:01:28.475
My background actually is
15
00:01:29.640 --> 00:01:35.420
in math finance. I wrote my PhD thesis, oh, already back in 2000. I finished my PhD,
16
00:01:35.935 --> 00:01:38.995
actually, and I've been working in the industry since then.
17
00:01:40.015 --> 00:01:45.840
So my my focus today is actually that we're working, with Python for finance, actually.
18
00:01:46.320 --> 00:01:48.659
I founded the company called the Python Quants,
19
00:01:48.960 --> 00:01:50.899
and there we focus actually on supporting,
20
00:01:52.079 --> 00:01:56.065
small, medium size, as well as the biggest institutions that we have in the financial industry,
21
00:01:56.445 --> 00:01:59.025
like hedge funds, big banks, exchanges,
22
00:01:59.645 --> 00:02:03.505
other data service providers, and so forth. So we're also, doing
23
00:02:03.940 --> 00:02:09.080
conferences in this in this context and then much, much more like running meetup groups in
24
00:02:09.380 --> 00:02:12.675
New York and London, actually. So it's quite diverse, but, actually,
25
00:02:13.875 --> 00:02:20.935
the common field is Python for finance. This this this is what you can say what I'm doing all day, all all week, all year, actually.
26
00:02:22.480 --> 00:02:23.780
Very cool. So,
27
00:02:24.320 --> 00:02:26.340
Ios, how did you get introduced to Python?
28
00:02:27.200 --> 00:02:32.405
Well, this is actually the incident. I was working for a technology client. It's, well,
29
00:02:32.865 --> 00:02:33.365
back
30
00:02:33.665 --> 00:02:34.885
more than 10 years.
31
00:02:35.825 --> 00:02:36.565
Back then,
32
00:02:36.945 --> 00:02:42.590
I was working for this technology client in a consulting capacity. I wasn't involved in technology itself,
33
00:02:42.970 --> 00:02:49.775
but there was this particular project where I learned to know Python. And I thought, well, this might be actually good language to do quantitative finance.
34
00:02:50.235 --> 00:02:50.974
Back then,
35
00:02:51.995 --> 00:02:54.815
maybe only a few have been using Python in this regard.
36
00:02:55.194 --> 00:03:13.035
When I started using it and I talked to other people about it, they always say, well, Python is a very well suited to attack the typical problems of quantitative finance because it's an interpreted language. It might be too slow, or they even say, well, it is too slow. You can't use it. But I think over the last 10 years, we have come a long way in the Python world.
37
00:03:13.990 --> 00:03:17.290
All these strategies, I would say, they are no longer
38
00:03:17.590 --> 00:03:37.810
true or valid these days. We have the amazing Python ecosystem that make Python a very good choice. So but again, it was just by incident. I like the language from the first day that I saw it. I thought it's so close to the mathematical syntax that you used to use in in, in quantitative finance. I said, I I thought simply this must be a perfect match, and I guess,
39
00:03:38.450 --> 00:03:40.230
history, at least the last 10 years,
40
00:03:41.045 --> 00:03:43.465
have provided lots of support, for my,
41
00:03:44.004 --> 00:03:53.760
back then it was just a vision or it was just like a feeling that this might be a good fit. But the biggest institutions in the world like Bank of America Merrill Lynch, they're heavy Python users these days. They prove right,
42
00:03:54.299 --> 00:03:56.799
that Python is indeed a very good choice for finance.
43
00:03:57.605 --> 00:03:59.944
Yeah. It's definitely pretty amazing how
44
00:04:00.325 --> 00:04:04.105
quickly the ecosystem has grown up around the use cases for doing
45
00:04:04.405 --> 00:04:11.720
quantitative analysis and various forms of data science and other kinds of number crunching within the Python language and all the different,
46
00:04:12.500 --> 00:04:19.535
sort of native libraries that are that have bindings to help alleviate that issue with speed that you mentioned.
47
00:04:20.955 --> 00:04:27.050
So can you explain what quantitative finance is and also perhaps differentiate that from algorithmic trading?
48
00:04:28.470 --> 00:04:33.370
You know what? Quant finance, I, at least as far as my thinking goes, I see
Load full transcript
49
00:04:33.815 --> 00:04:34.875
3 major subdisciplines
50
00:04:35.255 --> 00:04:36.475
under quantitative finance.
51
00:04:36.855 --> 00:04:41.035
There are, of course, those people who work on the models, I would say, for example, to value
52
00:04:41.440 --> 00:04:42.740
complex, derivatives
53
00:04:43.120 --> 00:04:45.060
like options or hybrid,
54
00:04:45.440 --> 00:04:51.575
securities or or other structures where you might have embedded options. These are the guys who write down the equations,
55
00:04:52.035 --> 00:04:53.655
who try to solve them by,
56
00:04:54.115 --> 00:04:55.975
mere frame power if you like.
57
00:04:56.515 --> 00:04:59.095
Then the second pillar from my point of view is computational
58
00:04:59.430 --> 00:05:03.530
finance where I say, well, now we have the model, but we have to implement that,
59
00:05:04.070 --> 00:05:19.300
by the means of for example, Python or typically what they use these days is c plus plus due to the speed and so forth. This is the second pillar, the computational part. And then, of course, we have, what I call a financial data science where this this whole financial logistics,
60
00:05:20.159 --> 00:05:25.699
topic that you need to cover in order to get your hands around the massive amount of data, actually.
61
00:05:26.825 --> 00:05:31.485
This is actually from my point of view, the closest part to algorithmic trading in that sense,
62
00:05:32.105 --> 00:05:34.845
that you have to process masses of of data.
63
00:05:35.620 --> 00:05:50.134
You can speak of big data in this regard, but, typically, I understand that the big data more or less, unstructured data. What we see in quantitative finance is actually usually very structured data, which you can store in tabular form. So there's no need to attack these,
64
00:05:51.150 --> 00:05:52.530
let's say, very structured
65
00:05:52.910 --> 00:06:13.500
datasets, which might be big with the typical means that are applied, for example, by Google or the Twitters of the world, where the problems are completely different. Speaking for example of the of the retail banking side, you have problems similar to the ones that are faced by Twitter and Google. But on the quantitative side, it's usually very structured data. Just think of kind of a table, a SQL,
66
00:06:14.040 --> 00:06:18.975
based table or, for example, a simple Excel spreadsheet where you have your numbers in such
67
00:06:19.354 --> 00:06:20.095
a 2 dimensional
68
00:06:21.035 --> 00:06:25.775
spreadsheet form. This is data that we face in quantitative finance. But nevertheless, the financial logistics
69
00:06:26.180 --> 00:06:27.640
part is, equally,
70
00:06:28.100 --> 00:06:29.880
as important as in the other disciplines.
71
00:06:30.340 --> 00:06:30.840
So,
72
00:06:31.300 --> 00:06:39.514
again, I see these 3 pillars, the theoretical part, the computation finance part, and the financial data science part, if you like. At least this is my thinking.
73
00:06:41.255 --> 00:06:51.950
And you alluded to it a little bit, but I'm curious how common it is for Python to be used in an investment bank or a hedge fund for the purpose of quantitative finance or possibly other uses as well.
74
00:06:53.755 --> 00:06:54.635
Well, I think,
75
00:06:55.755 --> 00:06:56.574
coming from,
76
00:06:57.355 --> 00:07:07.110
yeah, a history where Python hasn't played a role in the quantitative finance part, We usually see if finances is if Python is used at all kind of a blend of technologies,
77
00:07:07.889 --> 00:07:09.510
that they use in quantitative finance.
78
00:07:10.685 --> 00:07:11.664
I said it before,
79
00:07:12.445 --> 00:07:22.020
major banks these days have implemented the analytics libraries, their derivatives analytics, their risk analytics libraries in c plus plus due to the speed advantage that, these,
80
00:07:23.140 --> 00:07:24.440
compiled languages provided,
81
00:07:25.380 --> 00:07:27.640
10, 20 years ago, actually.
82
00:07:28.260 --> 00:07:29.000
But again,
83
00:07:29.384 --> 00:07:35.625
you have to differentiate this, to maybe new startup funds. Like, there are many, many hedge funds that you can get started,
84
00:07:36.185 --> 00:07:36.925
in 2015
85
00:07:37.465 --> 00:07:40.720
where many big shops, are closing. The funds actually.
86
00:07:41.740 --> 00:07:44.060
This this is 1 of the major headlines this,
87
00:07:44.700 --> 00:07:46.240
this year about the hedge funds.
88
00:07:47.155 --> 00:07:51.975
And these new shops, they they saw the potential for Python to provide a unified
89
00:07:52.355 --> 00:07:53.495
technology platform,
90
00:07:54.170 --> 00:08:04.095
in a sense that Python might not be only used for prototyping purposes or to script something up quite quickly, but also for kind of the core technology and the core,
91
00:08:04.474 --> 00:08:18.800
analytics speed, risk analytics, derivatives analytics, or for the purpose of implementing trading strategies and so forth. So it is used, but I think to a different degree depending on, on the, the companies you're looking at. I mentioned the example before bank of America Merrill Lynch,
92
00:08:19.180 --> 00:08:46.735
their their major, trading and risk management platform, which is called Quartz. The last number I heard is that they have about 12, 000, 000 lines of Python code in production there though. It might be 1 of the biggest financial applications in the world and they're obviously heavy, Python users. Others might be using it only to a smaller extent but again, all the big institutions these days be it hedge funds, be it asset managers or be it the big investment banks, they're more or less using Python but again to a different degree maybe.
93
00:08:47.980 --> 00:08:48.480
And
94
00:08:49.020 --> 00:08:55.360
just to dig a little bit deeper into the factors that contribute to the choice of whether or not to use Python in that kind of environment,
95
00:08:55.885 --> 00:08:56.785
I know you mentioned,
96
00:08:57.485
--> 00:09:01.825
having it as more of a unifying language, but I'm curious if you can just dig a bit more into that.
97
00:09:02.819 --> 00:09:03.720
So the unification,
98
00:09:04.259 --> 00:09:07.079
power of Python, I think, is 1 that is,
99
00:09:07.699 --> 00:09:12.600
yeah, dominant these days. So when you when you start out, then you can decide on your own. Again, the example,
100
00:09:13.205 --> 00:09:18.185
a team coming out of a big investment bank, they set up shop with their new fund, their emerging managers,
101
00:09:18.485 --> 00:09:55.775
they can decide on what technology to start with. Obviously, for a big investment bank, this is not a choice when you say, well, I have millions of lines of code, in c plus plus. There's no point, no business case in in migrating that to Python. But if you start out a new, I guess this is kind of, the right thing to go these day to use Python. Many, many do that, these days, actually. But there are other factors as well. I mean, it is kind of a a hip language these days, so many people like to work with Python. And I guess it's also a factor when it comes to recruiting. They just say, well, we are we are innovative. We use Python. You can work in Python here. They are not the old compiled languages that they need to use, that they need to master.
102
00:09:57.610 --> 00:09:58.270
But again,
103
00:09:58.650 --> 00:10:03.290
usually, it's kind of a blend of technologies, and they seldom people these days are writing,
104
00:10:03.610 --> 00:10:07.925
stuff only in 1 language. But Python might be a very good 1 when it comes to,
105
00:10:08.725 --> 00:10:09.225
fascinating
106
00:10:10.084 --> 00:10:13.545
and acquiring people and get the best recruits out there in the market, actually.
107
00:10:14.010 --> 00:10:21.310
Others, I mentioned that before, other factors, might be the ecosystem, of course. Things like pandas, data analytics library
108
00:10:21.695 --> 00:10:26.675
written back then or still supported by Wes McKinney. It's kind of a major library
109
00:10:26.975 --> 00:10:36.329
where I see that people get drawn into Python solely by this particular library. So, when I give seminars or or we're doing or running our conferences, they are
110
00:10:36.785 --> 00:10:46.010
very often people who come from a different background like r or MetLab and they say, well, I'm so fascinated by by what Python has to offer these days. But again, think back 5 years,
111
00:10:46.490 --> 00:11:01.455
we have come a long way when it comes to these libraries like pandas or performance library libraries like Python and Numba, which make, Python really fast. So I think we have lots to offer and this might be in the end, I guess, the the most important thing that we have this particular,
112
00:11:02.120 --> 00:11:04.139
powerful particularly powerful ecosystem,
113
00:11:04.680 --> 00:11:07.100
which I guess distinguishes Python, for example,
114
00:11:08.839 --> 00:11:27.410
from languages like Ruby, which are which is pretty close, in terms of syntax, which is also quite nice. Many people like it. But the ecosystem for doing what we do in quantitative finance is mainly missing in most other languages that you might come up with, when it comes to the beauty of the language itself, but not the powerful or the power of the the ecosystem,
115
00:11:28.085 --> 00:11:29.545
which surrounds the language.
116
00:11:31.125 --> 00:11:35.145
This is definitely something that I think the Python community should consider marketing
117
00:11:35.605 --> 00:11:41.290
a little bit more aggressively in that. I've actually had this discussion a number of times of late.
118
00:11:41.670 --> 00:11:44.490
You know, Python is 1 of those languages that can tend to be polarizing
119
00:11:45.175 --> 00:11:53.515
in terms of if people have really strong opinions about the, you know, syntactic white space or whatever the case may be. And I've just been trying to tell people,
120
00:11:53.950 --> 00:11:57.410
I don't care. Like, even if you hate the language itself,
121
00:11:58.030 --> 00:12:03.810
you know, it does the ecosystem that has been built up around it deserves to be explored because
122
00:12:04.335 --> 00:12:16.160
people are doing really interesting things with Python that just aren't happening in other programming language camps, at least not in the depth and diversity that they're happening here. Yeah. I I completely agree. I mean,
123
00:12:16.800 --> 00:12:17.779
if you differentiate
124
00:12:18.240 --> 00:12:21.565
the the work of a of a developer who's writing, like, production code,
125
00:12:22.525 --> 00:12:26.705
to the work, for example, of quant analyst who is more or less doing interactive
126
00:12:27.245 --> 00:12:27.745
work.
127
00:12:28.285 --> 00:12:31.400
And then you can say, for example, that pandas in and of itself,
128
00:12:31.780 --> 00:12:33.240
but represent kind of
129
00:12:33.700 --> 00:12:38.440
a new even higher language in a certain sense because you can work with pandas alone these days,
130
00:12:38.905 --> 00:12:48.509
which wraps actually all the nice libraries that are available in the ecosystem like, Py tables for HDF5 storage or map lib for plotting or num expression for parallel, numerical
131
00:12:49.610 --> 00:12:53.790
expression evaluation, and so forth. And you can, for example, if you're a non
132
00:12:54.175 --> 00:13:14.615
computer scientist or non developer whatsoever, you can learn pandas as a language in itself, not even taking too much care of what Python has to offer. So I I'm playing with this, thought for quite a while. I'm thinking of kind of, more like executive level, of people who might be interested in learning pandas alone. So what you're saying is is completely right. Even if you if you don't,
133
00:13:15.714 --> 00:13:31.355
like the language itself, the ecosystem around that and pandas from from my point of view is kind of the the prime example in this case. It's so powerful and and
you find so many useful things that you might not find in other areas that it's always worth to have a look at least at at at the specialized libraries in this regard.
134
00:13:32.135 --> 00:13:35.835
That's very interesting that you mentioned people coming to Python
135
00:13:36.370 --> 00:13:36.870
the
136
00:13:37.690 --> 00:13:38.190
language.
137
00:13:39.010 --> 00:13:39.510
And
138
00:13:43.570 --> 00:13:53.204
it reminds the language. And it reminds me a lot of the experience that people have had with Ruby and Ruby on Rails where people come to the Ruby community just to use Rails and
139
00:13:53.520 --> 00:13:58.740
for the first while that they're using it don't even necessarily realize that Ruby is its own distinct entity
140
00:13:59.120 --> 00:14:00.900
apart from that framework. So
141
00:14:01.245 --> 00:14:07.345
it's it's an interesting parallel to have that happening with pandas and using that to bring people to Python.
142
00:14:08.310 --> 00:14:14.330
Yeah. Exactly. Exactly. I mean, usually people have problems, and they want to solve their problems. And,
143
00:14:15.350 --> 00:14:15.850
only
144
00:14:16.265 --> 00:14:21.485
the second instance they care about the means in a sense that they are not religious about that. So
145
00:14:21.865 --> 00:14:24.045
again, the example is pandas
146
00:14:24.940 --> 00:14:40.285
has been modeled more or less in the beginning after the data frame, which is which is provided by r. But more and more people are coming to Python to to pandas these days because, the data frame that we have in pandas is much more powerful than the data frame that is available in R. So it's kind of,
147
00:14:40.685 --> 00:14:58.795
the upside down if you like that people now are coming to us because we have something which was modeled in the 1st place after something that the our community had available for quite a while. So, this is, also illustrative from my point of view for for the long way we have we have come and what we have to offer these days also for people, yeah, with a with a different,
148
00:14:59.195 --> 00:15:03.130
background and and, yeah, a different education in this regard, actually.
149
00:15:04.550 --> 00:15:10.410
So, Yves, are there any performance bottlenecks or other considerations inherent in using Python for quantitative finance?
150
00:15:11.625 --> 00:15:14.605
Yeah. Of course. I'm always saying, you can write
151
00:15:15.065 --> 00:15:17.920
really slow Python code. This is this is for sure.
152
00:15:18.560 --> 00:15:22.660
And there are many, many straps that you can tap into. The problem with Python,
153
00:15:23.839 --> 00:15:25.220
is that you can achieve
154
00:15:27.005 --> 00:15:36.350
the same result by, I don't know, a multitude of different ways because we have multiple paradigms that you can use. You have different idiomatic ways that you can choose,
155
00:15:36.730 --> 00:15:40.590
to reach the result and so forth. So for many people coming to Python,
156
00:15:41.425 --> 00:15:49.685
with a different background, it might be a little bit tricky to get it right, in the first place. For example, the major example is that many many algorithms
157
00:15:50.009 --> 00:15:50.589
in qualitative
158
00:15:51.130 --> 00:15:53.529
finance, when implemented in any kind of,
159
00:15:54.009 --> 00:15:54.509
language,
160
00:15:55.449 --> 00:15:59.835
programming language, are based in the end of of a loop or even a nested loop.
161
00:16:00.395 --> 00:16:23.084
And everybody knows that, Python loops are slow. And therefore, some are reasoning that, well, part of the finance is in the end about looping or nested looping, and Python is bad at looping. So Python must be bad for quantitative finance. But, again, when you write the programs and the algorithms right, you will, try to avoid writing loops on the Python level as far as possible. For example, using
162
00:16:23.399 --> 00:16:24.459
the powerful vectorization
163
00:16:24.839 --> 00:16:26.220
capabilities of NumPy,
164
00:16:27.160 --> 00:16:29.399
where the looping is so to say,
165
00:16:29.879 --> 00:16:31.740
delegated to the c level.
166
00:16:32.315 --> 00:16:50.185
So so to say in a machine room of NumPy, actually, where everything happens at the speed of c code because NumPy is more or less completely implemented in c. There's also a 4th round part, but usually what you're using as a default is the is the c part of the implementation of NumPy of the n d array object. And this is then in the end quite,
167
00:16:50.725 --> 00:17:14.535
quite fast actually, but you have to know that. People coming from r or for example, MATLAB, they might know that. But people coming, for example, with the c plus plus or c sharp backgrounds or or Java where they used to write their, their loops. Yeah. They might be inclined to to write their loops in Python as well. And this might be, for example, a track where you can tap into where you say, well, performance is is all about what I what I expected or what we need here for this particular application.
168
00:17:16.060 --> 00:17:23.680
And this might be tricky in the 1st place, but I guess a good training or there are so many tutorials available on the web to teach that stuff. This might
169
00:17:24.380 --> 00:17:36.190
get you around the the traps that I was mentioning before. And in the end, if this is not enough, let's say for example, working with the umpire is nice, but my my special problem cannot be cast into a setting where full vectorization
170
00:17:36.650 --> 00:17:37.470
or for example,
171
00:17:37.770 --> 00:17:40.350
I cannot afford to allocate too much memory,
172
00:17:41.530 --> 00:17:48.945
for the vectorization. Then there are other libraries I mentioned, for example, Numba before, which might help you, for example, with your with your specialized,
173
00:17:49.404 --> 00:17:53.985
loops and asset loops. And it's a very powerful tool to, dynamically compile
174
00:17:54.429 --> 00:18:03.090
your pure Python code if you like to a machine code, and then you get a speed again, but in a more or less than customized fashion. So there are
175
00:18:03.405 --> 00:18:10.225
bottlenecks if you like, if you're doing it wrong. But usually, if you know how to do your stuff for qualitative finance, for computational finance,
176
00:18:10.720 --> 00:18:15.600
then Python might be as fast as any other language that, you use usually for these,
177
00:18:16.000 --> 00:18:16.980
particular topics.
178
00:18:17.519 --> 00:18:18.580
You asked before
179
00:18:19.034 --> 00:18:25.294
about algorithmic trading or if you think for high frequency trading. I mean, when we speak of the milliseconds that matter,
180
00:18:25.595 --> 00:18:29.519
in these particular areas, then Python might indeed not be the right choice. So,
181
00:18:29.980 --> 00:18:33.919
I always differentiate when I say, well, for the typical, big data problems,
182
00:18:34.620 --> 00:18:37.039
Python is hasn't been developed actually,
183
00:18:37.375 --> 00:18:44.654
also not for the for the faster things in the world, but Python can interact with all the technologies that are used there, both in a big data,
184
00:18:44.975 --> 00:18:46.115
world and in the high frequency and then the
185
00:18:46.430 --> 00:18:48.210
and then the in the high performance,
186
00:18:49.070 --> 00:18:58.394
area if you like. But Python in and of itself was never designed to attack these particular problems. But again, as a general language, as a general programming language attacking
187
00:18:58.695 --> 00:19:02.235
the regular problems, let's say that the 98% of typical problems,
188
00:19:02.550 --> 00:19:12.955
Python might be, actually a good choice to go for. But again, you have to take care of the idioms of the approaches that they use because otherwise, you might end up indeed with slow, nonperforming Python code.
189
00:19:13.815 --> 00:19:25.980
That's really interesting that you and and a few other people also have brought up the slowness around loops and and iteration in Python. I wonder if in, you know, Guido and and all the other folks sort of, you know, working on python.next
190
00:19:26.600 --> 00:19:32.115
and and and investigating the future of the language, I wonder if any thought is being given towards
191
00:19:32.575 --> 00:19:33.875
how that performance
192
00:19:34.655 --> 00:19:36.115
issue might be mitigated.
193
00:19:37.760 --> 00:19:40.740
I think, for sure, they are working on that. But by the very design,
194
00:19:41.280 --> 00:19:55.095
there are limits to solving this particular issue. First of all, Python is a dynamically typed language, so the interpreter has to decide during runtime what kind of type it faces. This brings a lot of generality, but also, obviously, lots of overhead,
195
00:19:55.570 --> 00:19:59.990
compared to a compiled language where you say, well, this is a float. This is that. This is this.
196
00:20:00.290 --> 00:20:12.434
In Python, you can
pass kind of anything, for example, into a function, and then the interpreter has to decide what has passed to the function. And for example, there might be functions that work on a float, that work on an array, and they might even also work on a string object.
197
00:20:12.780 --> 00:20:20.145
This is the nice point about that, about dynamic typing that you get lots of generality, but this brings again, the overhead with it.
198
00:20:20.625 --> 00:20:26.245
On the other end yeah. I mean, it is and is, an interpreted language. And if you speak of the CPython
199
00:20:26.545 --> 00:20:27.045
implementation,
200
00:20:27.510 --> 00:20:39.085
this is as it is. Of course, we have, the other Python versions like the Python project where you have something like just in time compiling. And then we usually see speed ups of like 6 to 8 times, compared to
201
00:20:39.625 --> 00:20:40.649
the compared to the,
202
00:20:41.465 --> 00:20:42.445
interpreted version,
203
00:20:42.825 --> 00:20:54.730
actually. So, obviously, there are projects implemented. But in the end, if you stick to the 2 major paradigms that have a dynamically typed language and you have an integrated 1, then there might be limits to speeding up, the the,
204
00:20:55.130 --> 00:20:58.225
the loops. But in the end, from my point of view for quantitative finance,
205
00:20:58.525 --> 00:21:05.470
what we are talking about, this is actually not a problem. But if you have if you have, for example, a a big application or, let's say,
206
00:21:06.429 --> 00:21:20.365
a module from a from a analytics library, then usually you have kind of 2 or 3 spots only where this bottleneck might arise arise where you say, well, here is kind of my nested loop, which is too slow. And then you can pick that particular part out,
207
00:21:21.060 --> 00:21:33.164
and maybe vectorize it as a first approach or you do dynamic compiling or you do a static compiling using and then interface with the with the compiled version and then the bottleneck, so to say, has vanished.
208
00:21:33.705 --> 00:21:39.030
You don't have to write, like, if you have a 1, 000 lines of code and maybe 5 or 10 lines only might
209
00:21:45.910 --> 00:21:49.425
and to rewrite or to compile, dynamically, statically compile
210
00:21:49.725 --> 00:21:58.370
the 5 or 10 lines of code which represent the bottleneck, and this is kind of my thing. You can start out with pure Python, your profile, your code, you say, well, here here we see kind of,
211
00:21:59.250 --> 00:22:10.600
where where the slowness comes from, the execution, and then you go you go on and and improve this particular part. In the end, you might be as fast as with kind of a completely compiled version of what you have
212
00:22:11.480 --> 00:22:17.340
given. And, just to go a little bit deeper into the weeds for profiling your code, do you just use the standard,
213
00:22:17.960 --> 00:22:23.965
built in profiling module from the standard library? Or are there any other tools that you like to leverage for that purpose?
214
00:22:24.424 --> 00:22:27.804
Yeah. Yeah. There are multiple options available. What we kind of use is kind of
215
00:22:28.265 --> 00:22:29.965
usually the very standard. But
216
00:22:30.450 --> 00:22:31.909
if you have experience,
217
00:22:32.210 --> 00:22:34.150
you usually don't need to,
218
00:22:34.610 --> 00:22:36.390
to run it through a profiling engine,
219
00:22:36.770 --> 00:22:39.830
to say, well, this is where the bottleneck is. The
220
00:22:40.505 --> 00:22:50.659
majority of algorithms in the end look like the same. And with a little bit of experience, you know where the where the where the important parts are, which you have to, yeah, compile in the end. So this is more or less kind of,
221
00:22:51.380 --> 00:22:54.840
more a thing of experience than of profiling in the end,
222
00:22:55.380 --> 00:22:57.320
I would say, actually, for what we do.
223
00:22:58.575 --> 00:23:04.195
So what kinds of background are typical and necessary for getting started in quantitative finance?
224
00:23:06.050 --> 00:23:08.550
I think here, we also have come a long way.
225
00:23:09.090 --> 00:23:12.870
Back then, I would say so in the in the eighties, even in the nineties,
226
00:23:13.875 --> 00:23:21.575
what they call the people who did qualitative finance, they were usually called the rocket scientists, and and the major reason for this was because they have been rocket scientists,
227
00:23:21.890 --> 00:23:22.390
actually.
228
00:23:22.850 --> 00:23:27.670
Many people had their background in physics or mechanical engineering or in mathematics
229
00:23:28.370 --> 00:23:32.955
or or some similar very formal field where they got the the necessary
230
00:23:34.055 --> 00:23:36.475
math education to attack quantitative finance
231
00:23:37.095 --> 00:23:41.889
problems. These days, when you have a look at the educational market, you have many, many specialized degrees,
232
00:23:43.389 --> 00:23:45.649
even called the master of financial engineering,
233
00:23:45.950 --> 00:23:47.649
master of quantitative finance.
234
00:23:48.535 --> 00:23:55.435
So again, we've come a long way in a sense that we have today the the the education industry, the universities that support
235
00:23:56.240 --> 00:24:00.900
education in this regard. And I guess people interested in getting into quantitative finance,
236
00:24:01.440 --> 00:24:07.205
they usually go that route that they do their bachelor and and later on get their specialization with the typical program.
237
00:24:07.905 --> 00:24:10.725
I don't know how many there are alone in New York and in London,
238
00:24:11.505 --> 00:24:16.470
where they get their master or even later on they go on, like, the PhD thesis that she got.
239
00:24:16.850 --> 00:24:18.309
So this is kind of the background,
240
00:24:18.769 --> 00:24:22.549
that is typically, I would say, these days required that you know how to program.
241
00:24:22.904 -->
00:24:27.005
They learn, for example, c plus plus and more and more Python these days, obviously.
242
00:24:28.105 --> 00:24:34.880
And, they get the the math education that they need to to work in this particular field. But, obviously, there are other people
243
00:24:35.420 --> 00:24:38.560
that might come with, for example, a math background or an engineering background,
244
00:24:38.860 --> 00:24:43.965
and they are already in their professional careers, and they wanna get into the field. Then there's also,
245
00:24:44.524 --> 00:24:46.304
a professional education market.
246
00:24:46.845 --> 00:24:50.544
For example, Fitch Learning, they offer kind of the very famous, CQF,
247
00:24:50.990 --> 00:25:02.125
the certificate in quantitative finance. I think since more than 12 years, I, this year, got a lecturer for the program. So they introduced Python as well. Here we see that Python is scanning crawling in this part,
248
00:25:02.525 --> 00:25:04.225
of the spectrum as well,
249
00:25:04.605 --> 00:25:05.105
because,
250
00:25:05.565 --> 00:25:17.720
Raveed Cook, which is the academic director, he said, well, we have to keep up with the times and Python obviously has become a force in quantity finance. So we wanna introduce Python into the curriculum and and here we go. So I'm teaching,
251
00:25:18.485 --> 00:25:22.664
quant finance or actually computational finance. I must I must specify given my own,
252
00:25:22.965 --> 00:25:23.465
subclassification,
253
00:25:24.565 --> 00:25:31.899
based on Python in this program right now. And I guess going forward, more and more Python will also be included in this
254
00:25:32.679 --> 00:25:35.625
professional education program. We also, for example,
255
00:25:35.925 --> 00:25:41.545
discussing, and we will provide this probably next year for the first time to do a certification
256
00:25:42.440 --> 00:25:54.445
in Python for quantitative finance, which is, a little bit shorter, but a little bit more specialized. So for people who say, well, I might have a background, for example, in financial engineering with masters, but I've never done any Python.
257
00:25:54.985 --> 00:26:07.320
But I know for example what the plexigold, stochastic differential equation is for the geometric Brownian motion, then they might learn during this certification, which will be, over 4 days with the with kind of final test, a final exam,
258
00:26:07.780 --> 00:26:10.915
then would then they will learn how to apply Python actually,
259
00:26:11.535 --> 00:26:17.295
in in this particular field. So there are many, many ways. Usually, these days, I guess, people who are interested, they they start out,
260
00:26:17.855 --> 00:26:19.155
studying this at university.
261
00:26:19.550 --> 00:26:21.570
But again, there's also kind of a market,
262
00:26:22.430 --> 00:26:24.130
which provides professional
263
00:26:24.670 --> 00:26:25.170
education
264
00:26:25.630 --> 00:26:32.135
offerings for those who are already in their
career and now want to shift focus or want to deepen their knowledge in this particular field.
265
00:26:33.635 --> 00:26:43.980
Yeah. It's interesting how new technologies have given rise to completely new classes of occupation that didn't exist before computers became so ubiquitous and also how
266
00:26:44.600 --> 00:26:46.380
software in general has
267
00:26:46.775 --> 00:26:48.475
allowed a number of people to
268
00:26:49.015 --> 00:26:50.315
reimagine their careers
269
00:26:51.015 --> 00:27:03.100
at various points without necessarily having to go all the way back to school and get another degree or take a significant amount of time off of the work that they're already doing in order to gain new certifications.
270
00:27:03.785 --> 00:27:08.605
So what kinds of libraries or algorithms in Python are useful for the day to day work of a quant?
271
00:27:10.025 --> 00:27:17.570
Well, actually, this is pretty similar to what data scientists use. I mean, data scientists might be a little more popular field for which Python is used.
272
00:27:18.029 --> 00:27:26.264
But we in Quant Finance use more or less the same toolset. So it all starts out with the with the scientific stack as it's usually called,
273
00:27:26.804 --> 00:27:32.990
which is based more or less completely on NumPy, which gives the, and the array object for numerical computations
274
00:27:33.290 --> 00:27:34.670
based on array structures.
275
00:27:35.610 --> 00:27:42.005
Then obviously, I mentioned it quite a few times already, using pandas for data crunching, for visualization for example.
276
00:27:42.865 --> 00:27:48.360
And these are already the 2 major libraries I'm working with on a day to day basis.
277
00:27:48.820 --> 00:27:56.305
Of course, there are much more specialized libraries and so forth, but I guess these 2 are the workhorse actually these days.
278
00:27:57.025 --> 00:28:01.925
I really like, storing data in HDF 5 format, which is a very fast format,
279
00:28:02.385 --> 00:28:07.050
binary format to store data almost at the speed of the hardware, that is available.
280
00:28:07.590 --> 00:28:17.115
So usually even working on a on a small MacBook that I have in front here, I get to, write and read speeds of 500 megabytes per second with Python is already built in.
281
00:28:17.495 --> 00:28:26.919
This is my example that I'm usually mentioning when I say, well, the the HPC part when it comes to IO is usually built in in Python. Other language, they have to,
282
00:28:27.460 --> 00:28:30.360
yeah, crap a little bit more with the input output
283
00:28:31.495 --> 00:28:35.115
part. And there are many others. There's so many so many nice,
284
00:28:35.655 --> 00:28:38.670
libraries and stuff out there, but I think in the end, it
285
00:28:39.230 --> 00:28:41.090
very problem that that you are attacking.
286
00:28:41.630 --> 00:28:49.255
But again, I can only stress the 2 ones that are the workhorse. It's it's, NumPy and it's Pandas. And I mentioned already
the the the,
287
00:28:50.035 --> 00:28:50.535
performance,
288
00:28:51.155 --> 00:29:15.340
libraries, which might be useful. But this more or less on a case by case basis that you might wanna use number for example when you when you face a nested loop which cannot be vectorized or which you wanna implement in a memory efficient way and all the the other nice things that we have like Thyssen and so forth. So but again, it is and stays the the scientific stack, which is used in almost any kind of scientific discipline or the data science part.
289
00:29:16.039 --> 00:29:24.455
And quantitative finance, at least in this regard, is not that different from the other application areas of Python and the and the ecosystem surrounding it.
290
00:29:25.635 --> 00:29:31.750
So talk me a little bit more about the actual data that is manipulated and used for powering these decisions. I'm curious,
291
00:29:32.450 --> 00:29:41.015
I guess, where the different sources are. So is it all just market financial data? Is there proprietary data that companies bring into play? And also,
292
00:29:41.955 --> 00:29:47.360
how useful or prevalent is semantic analysis of various news sources when
293
00:29:47.740 --> 00:29:52.240
trying to reach decisions about what sorts of trades or investments to make?
294
00:29:53.965 --> 00:30:10.820
1 part I measure also again before is kind of that we usually on the at least on the quantitative finance side, we're very structured data, and the structured data is typically kind of a financial time series Mhmm. Which says you have your day time stamps, let's say, on the on the 2nd March last year at 11 o'clock.
295
00:30:11.345 --> 00:30:16.245
We had, for example, an index level for the S and P 500 of x y zed.
296
00:30:16.865 --> 00:30:24.980
And this then you have, for example, over 30 years, you might have this on a daily basis or on a minis bar basis or even on the second level.
297
00:30:25.424 --> 00:30:26.085
But nevertheless,
298
00:30:26.385 --> 00:30:29.525
the structure of the data is the same. It's typically financial time series.
299
00:30:30.705 --> 00:30:35.450
And these financial time series are generally provided by, at least in the professional context,
300
00:30:35.750 --> 00:30:39.769
by the professional data, providers like Bloomberg and Thomson Reuters,
301
00:30:40.149 --> 00:30:43.585
which, by the way, both of both of them provide Python,
302
00:30:43.985 --> 00:30:48.245
based wrappers around their APIs. So it's very very convenient usually,
303
00:30:48.784 --> 00:30:51.924
to interface with the day to day provider from a Python level.
304
00:30:52.240 --> 00:30:55.220
Actually, it's not where it could be or should be.
305
00:30:56.320 --> 00:30:57.700
I heard many people complaining
306
00:30:58.160 --> 00:31:03.255
about practical issues in this regard, but at least there is the the possibility to interface from Python
307
00:31:03.555 --> 00:31:04.375
to these sources.
308
00:31:04.755 --> 00:31:09.015
But there are other areas as well where you can get financial data, and this is kind of a very nice
309
00:31:09.740 --> 00:31:11.600
development I see in our market,
310
00:31:12.140 --> 00:31:25.335
that we not only have open source, and we have the GitHub, and the Meetup movement, and you name it. We also have open data, which is coming more and more. And 1 of, the bigger players in this regard in our field, in the financial field is Quanda. It's called Quanda,
311
00:31:25.794 --> 00:31:28.294
which provides kind of a unified interface
312
00:31:29.390 --> 00:31:32.450
to I don't know how many open data sources actually.
313
00:31:33.790 --> 00:31:55.305
For example, 1 of my examples I I use on on our platform on Datapark. Io is interfacing with Quandl and analyzing Bitcoin historical data for with regard to the exchange rate to the US dollar, with regard to the number of Bitcoins issued, with regard to the volume of Bitcoins traded, but they and so forth. So this is this is kind of an example of what kind of data you get there. You also get data from Quanda,
314
00:31:55.685 --> 00:31:57.925
for free in an open standardized fashion,
315
00:31:59.100 --> 00:32:02.560
for historical data. For example, for indices, for
316
00:32:03.020 --> 00:32:03.520
commodities,
317
00:32:03.980 --> 00:32:17.600
for foreign exchange rates, and so forth. But not only that, there are wealth of of different data source that you can tap into there. So if you have never heard of it, just give it a try. Have a look at it as a standard rest rest API,
318
00:32:17.980 --> 00:32:23.279
JSON base, and you can easily interface, with Quandl, not only obviously from Python, but with any language,
319
00:32:23.740 --> 00:32:33.375
because it's it's such a nice API that they provide. And this is kind of a major trend, which is, really useful. When I think back when I wrote my my diploma thesis and my PhD thesis,
320
00:32:33.830 --> 00:32:34.230
getting,
321
00:32:34.870 --> 00:32:37.530
even small chunks of data was was really
322
00:32:37.910 --> 00:32:42.425
painful or in the end costly. So I remember 1 instance where I had to pay my professor like
323
00:32:42.985 --> 00:32:46.845
back then when I studied 20 Deutschmarks for a very small,
324
00:32:47.305 --> 00:32:56.300
data set that I needed. So it was much money back then for myself actually. But these days, when when you study in this field, you have so many open data sources that you can tap into.
325
00:32:56.760 --> 00:33:18.705
I think the most popular 1 because it's around since quite a while is the the Yahoo Finance data set where you can for example, from pandas quite easily interface with And if you're, for example, interested in the in the stock quotes of German Daimler Benz or famous car manufacturer or for the German DAX index or to this and for the mix volatility index in the United States, you get the data for free,
326
00:33:19.245 --> 00:33:20.625
from Yahoo, for example.
327
00:33:21.245 --> 00:33:24.065
So, this is kind of a major trend that I see
328
00:33:24.730 --> 00:33:29.230
coming or becoming even more, force in the future that we not have only,
329
00:33:29.610 --> 00:33:34.350
open source, but also open data in that sense that people can, easily work with.
330
00:33:35.635 --> 00:33:36.835
With regard to,
331
00:33:37.155 --> 00:33:44.215
the other part of the question was with regard to semantic analytics. I have seen quite a few talks and I know quite a few people who are working on that.
332
00:33:45.380 --> 00:33:53.880
But I don't have a coherent opinion on that. This is not a field that we are working with. So news analytics is an active field.
333
00:33:54.355 --> 00:33:55.175
There are companies
334
00:33:55.635 --> 00:33:59.735
who exclusively focus on that like Ravenpack, for example, is 1 provider
335
00:34:00.115 --> 00:34:01.895
of, so to say, insights
336
00:34:02.355 --> 00:34:04.159
based on, semantic analysis.
337
00:34:06.240 --> 00:34:22.170
But again, this is neither a field, which I would consider 1 of our expertise nor do I have kind of a coherent view on on how far fetched it's used. I think all the big players in the world when it comes to the buy side, the hedge fund and asset managers, they have have tried it out, but I don't
338
00:34:22.630 --> 00:34:23.130
know
339
00:34:23.589 --> 00:34:31.050
how far they really base in the end their decision to trade on this kind of information. I've seen quite good research results
340
00:34:31.495 --> 00:34:42.849
actually with a percent of their their studies, but, in the end approval lies and using it, to make real money. And there, I I can't mention numbers. So I say, well, this, this or that shop has used it,
341
00:34:43.710 --> 00:34:48.695
to generate, I don't know, 3 percentage points or plus over kind of benchmark or whatsoever.
342
00:34:49.155 --> 00:34:52.195
So I'm not an expert in that. Maybe others, might,
343
00:34:52.675 --> 00:34:54.375
have a stronger views on that.
344
00:34:56.380 --> 00:34:56.880
So
345
00:34:57.340 --> 00:35:02.960
this next question that I had may potentially have been somewhat off the mark, and you'll forgive me if so.
346
00:35:03.974 --> 00:35:11.035
But in any case, is Python actually used to enact the trades? What protocols, APIs, and libraries are used in this process?
347
00:35:13.030 --> 00:35:17.610
The first part of the answer is Python actually used to enact the trades, I would say, for sure. Yes.
348
00:35:17.990 --> 00:35:18.470
But,
349
00:35:19.030 --> 00:35:19.690
I think
350
00:35:20.245 --> 00:35:26.585
depending again on the company you look at, it might be from completely different angles that the input,
351
00:35:27.765 --> 00:35:31.960
comes from the Python script or application or whatever is used there.
352
00:35:32.260 --> 00:35:33.540
So some might even,
353
00:35:33.940 --> 00:35:34.820
implement their,
354
00:35:35.615 --> 00:35:38.915
the complete, trade process based on on on Python.
355
00:35:39.215 --> 00:35:47.450
Others might only use it, for example, to analyze data, come up with an investment idea, and then implement it later on using a different technology.
356
00:35:48.150 --> 00:35:51.690
So, from the 1 end, doing it completely based on Python, like automatons,
357
00:35:52.345 --> 00:35:52.845
iometric,
358
00:35:53.545 --> 00:36:01.725
algorithmic trading to just like being a supportive tool in order to come up with ideas and to back up ideas, so to say.
359
00:36:02.340 --> 00:36:03.240
I can see,
360
00:36:04.420 --> 00:36:04.920
Python,
361
00:36:05.540 --> 00:36:11.454
yeah, used in many many places in in this particular regard. When it comes to protocols, APIs, and libraries,
362
00:36:12.155 --> 00:36:13.775
the first part is not that,
363
00:36:15.994 --> 00:36:18.174
not that easy to answer. I mean, there are
364
00:36:18.860 --> 00:36:22.880
so many different applications, so many different areas where where it is used.
365
00:36:23.340 --> 00:36:28.240
I wouldn't say that there is kind of a a very standard. What we see these days, obviously, is JSON based
366
00:36:28.684 --> 00:36:34.865
APIs. We can easily interact with, for example, online brokers, for example, where they have standardized APIs. But again,
367
00:36:35.405 --> 00:36:37.744
shifting from 1 online broker to the next,
368
00:36:38.140 --> 00:36:40.560
you might encounter a completely different API.
369
00:36:41.020 --> 00:36:42.880
So I don't see any kind of
370
00:36:43.180 --> 00:36:45.280
real standardization in this regard.
371
00:36:45.740 --> 00:36:46.880
When it comes to libraries,
372
00:36:47.895 --> 00:36:48.395
actually,
373
00:36:49.015 --> 00:36:49.915
they are also,
374
00:36:50.375 --> 00:36:52.215
at least when I give my talks at,
375
00:36:52.615 --> 00:36:53.835
Python conferences,
376
00:36:54.215 --> 00:37:00.829
what I mentioned usually usually is that this is kind of I wouldn't say a blind spot actually that we don't have financial libraries.
377
00:37:01.289 --> 00:37:01.950
What we
378
00:37:02.569 --> 00:37:03.390
don't have
379
00:37:03.690 --> 00:37:04.750
really standardized
380
00:37:05.130 --> 00:37:06.505
ones yet or,
381
00:37:07.365 --> 00:37:10.025
I'll say powerful enough libraries yet.
382
00:37:10.404 --> 00:37:16.990
We have the the amazing Python ecosystem we have been talking about already a few times with non Python and so forth.
383
00:37:17.450 --> 00:37:23.905
But there is not kind of when you say, well, which 1 is the financial library I should use when I wanna trade this or that?
384
00:37:24.385 --> 00:37:30.965
This is still missing. What we are working on since quite a while is a a standardized library which you call DX Analytics.
385
00:37:31.540 --> 00:37:35.800
You find all the information on the website just just Google it or go to dxdashanalytics.com.
386
00:37:38.339 --> 00:37:41.845
But this currently has a strong focus on derivatives analytics,
387
00:37:42.145 -->
00:37:50.085
which might not be the focus of the majority of people, who use Python for finance. But I think when it comes to derivatives and risk analytics
388
00:37:50.450 --> 00:37:51.270
and portfolio analytics,
389
00:37:52.130 --> 00:38:00.905
we might, have to offer some special things here. So this is also 1 of my focuses for next year to build out our library and discussion,
390
00:38:01.685 --> 00:38:06.425
right now with, a few bigger partners who might support, our actions in this regard,
391
00:38:07.160 --> 00:38:15.820
to build our DX analytics to more or less kind of a full fledged library covering also back testing, trading, more on the portfolio analytic sides, and and,
392
00:38:16.295 --> 00:38:22.715
including more models like, currently we're working, for example, on the LIBOR market model, which is kind of very complex 1,
393
00:38:23.230 --> 00:38:26.530
but heavily and often used when it comes to the interest rate space.
394
00:38:27.150 --> 00:38:28.829
So in this regard, we have been,
395
00:38:29.309 --> 00:38:36.675
too weak if you like. But again, on the other part, there's the the real estate analytics part, and there the focus is on equities.
396
00:38:37.535 --> 00:38:48.119
I think our library is already quite strong, and there it is also used kind of in practice for for some of our clients where we do kind of derivatives and portfolio analytics on a regular basis have implemented
397
00:38:48.555 --> 00:38:50.015
the parts of their analytics,
398
00:38:50.475 --> 00:38:53.695
infrastructure based on our open source library, DX Analytics.
399
00:38:54.555 --> 00:39:00.260
And there are others, of course, not only ours. There's, for example, the SIP line library, which was open sourced by Quantopian,
400
00:39:00.880 --> 00:39:02.420
which is an algorithmic platform,
401
00:39:02.800 --> 00:39:03.540
and the crowdsourced,
402
00:39:04.000 --> 00:39:05.140
hedge fund. Actually,
403
00:39:05.525 --> 00:39:15.150
they open source tip line, which has a focus on back testing trading strategies, so they are more on the trading side. And they just recently open sourced another library, which they call Pie Folio,
404
00:39:15.610 --> 00:39:17.770
which is more or less about the,
405
00:39:18.250 --> 00:39:18.750
performance
406
00:39:20.170 --> 00:39:29.395
estimation of different trading strategies. So the 2, the supply and the portfolio are closely intertwined. The 1 is for back testing and coming up with the trading strategies, and the other 1 is for,
407
00:39:30.194 --> 00:39:33.240
estimating the performance and giving lots of statistics
408
00:39:33.860 --> 00:39:36.600
for different trading strategies in practice.
409
00:39:37.780 --> 00:39:40.260
And this is actually what what their platform is about.
410
00:39:40.900 --> 00:39:41.720
We are providing
411
00:39:42.135 --> 00:39:42.955
people an environment
412
00:39:43.255 --> 00:39:44.715
to, to generate
413
00:39:45.015 --> 00:39:57.110
trading ideas and to test them in practice, either by paper trading or going direct to, interactive brokers where they can trade with, real money if they like. If they have enough trust in their in their trading strategies.
414
00:39:57.570 --> 00:40:02.935
So again, there are some financial libraries. Another 1 I'm at, I could mention is, actually,
415
00:40:04.275 --> 00:40:07.015
written by a friend of mine, Sayeed Amin.
416
00:40:07.474 --> 00:40:12.339
He has also focused more on the trading side and the data crunching side. He He has, for example, interfaces to Bloomberg.
417
00:40:12.880 --> 00:40:15.460
But, again, we we are missing kind of this
418
00:40:15.760 --> 00:40:19.380
this 1, this single library, and I imagine something like pandas.
419
00:40:20.545 --> 00:40:30.550
What what pandas has accomplished for this data science part, here kind of groundwork. So this covers kind of 89% of what we typically see on a daily basis, and I'm imagining something,
420
00:40:31.090 --> 00:40:37.109
here in the future. And, again, my my priority, and this is written here on my whiteboard to the right of my desk
421
00:40:38.195 --> 00:40:40.935
for 2016 is to build out our library
422
00:40:41.235 --> 00:40:59.025
as far as possible and and maybe to get to a standard and to benchmark library if you like. In this regard, we can say, well, Python also has kind of this 1 particular financial library where you can go to and you find kind of a multitude of useful things that you need to do, quantitative and and computational finance.
423
00:41:00.445 --> 00:41:09.800
Very cool. I wanna ask you a question about DX Analytics in particular. So, obviously, I have a very limited understanding of the intricacies of derivatives trading.
424
00:41:10.100 --> 00:41:12.965
I did sort of do a chunk of
425
00:41:13.505 --> 00:41:18.645
reading around the market crash of a few years ago that a lot of people sort of blame
426
00:41:18.945 --> 00:41:29.605
derivatives trading for taking a rather large part in the the American market crash at least. Does DX Analytics sort of help you analyze the actual worth of a given derivative
427
00:41:29.905 --> 00:41:35.205
or, like, does it help you try to make smart investment choices around, are these derivatives
428
00:41:35.880 --> 00:41:39.579
good investments or not? Like, what what kinds of things can you do with it?
429
00:41:40.119 --> 00:41:42.539
Well, I think it's not as simple as that, unfortunately.
430
00:41:42.839 --> 00:41:45.339
So a typical Derivatives analytics library
431
00:41:46.025 --> 00:41:56.510
has not the main objective, at least from my point of view, that you say, well, this is kind of a good derivative or not. It always depends on what you wanna do, what your expectations are, what your kind of base position is.
432
00:41:56.970 --> 00:41:57.470
Usually,
433
00:41:59.130 --> 00:42:02.829
these libraries only help in coming up with your evaluations,
434
00:42:03.915 --> 00:42:05.775
where your estimates if you
like,
435
00:42:06.635 --> 00:42:08.655
and also with kind of risk
436
00:42:09.675 --> 00:42:10.175
figures,
437
00:42:10.640 --> 00:42:13.619
the so called creaks. I mean, now we're getting a little bit technical,
438
00:42:14.240 --> 00:42:23.705
which you need to manage your derivatives position. In this regard, our library might help people because we're implementing kind of a rather new approach when it comes to front office analytics,
439
00:42:24.405 --> 00:42:26.985
to get a little bit of a more comprehensive
440
00:42:27.380 --> 00:42:28.280
picture of
441
00:42:28.580 --> 00:42:29.960
what the derivatives portfolio,
442
00:42:30.740 --> 00:42:31.240
might,
443
00:42:31.540 --> 00:42:33.480
be all about in a sense that,
444
00:42:34.020 --> 00:42:35.400
we provide much more
445
00:42:35.885 --> 00:42:42.205
statistical information because the whole library is based on is based on Monte Carlo simulation, which which allows you to,
446
00:42:42.925 --> 00:42:43.745
get a complete
447
00:42:44.380 --> 00:42:46.240
picture of the of the distribution.
448
00:42:46.540 --> 00:42:47.280
For example,
449
00:42:47.660 --> 00:42:57.725
to make it a little bit more concrete. So I don't wanna use too many technical terms, technically quant finance terms. What we do with the library, for example, if you have a if you think of a derivatives portfolio which consists of 10
450
00:42:58.185 --> 00:42:59.005
different options,
451
00:42:59.545 --> 00:43:00.525
complex derivatives,
452
00:43:01.680 --> 00:43:14.615
Then there might be kind of subtle interactions between the ones when, for example, the market moves when you're when you're trading, for example, options on the s and p 500 and so forth. What we do with the library is what is typically done in the back office these days,
453
00:43:15.155 --> 00:43:18.135
but we do it as a front office approach in a sense that
454
00:43:18.470 --> 00:43:23.609
we do multi color simulation for the front office analytics. And this allows us, for example, to get the complete distribution
455
00:43:24.390 --> 00:43:24.630
of,
456
00:43:25.510 --> 00:43:26.809
of, the
457
00:43:27.255 --> 00:43:28.695
possible future pay
458
00:43:29.495 --> 00:43:35.035
of such a complex portfolio in a sense that we're implementing, for example, 250, 000 or even a 1000000
459
00:43:35.360 --> 00:43:46.484
simulations for a complete portfolio. So we end again, ending up with not only kind of a single point estimate that we say, well, the portfolio has currently a worth of say, 1, 000, 000.
460
00:43:47.185 --> 00:43:53.285
We get kind of a distribution of the of the potential future values. And in this sense, actually,
461
00:43:54.160 --> 00:44:03.855
we get a more complete and more comprehensive picture when it comes to the analysis of these portfolios. But I wouldn't say that DX Analytics in the first place really helps you in deciding,
462
00:44:04.315 --> 00:44:08.095
which options to choose and so forth. This is driven more or
less either by a necessity,
463
00:44:08.795 --> 00:44:09.295
given
464
00:44:10.030 --> 00:44:16.690
the portfolio that you have already or by the simply the strategy you wanna implement. But when it comes to analyzing your positions and what what you
465
00:44:17.395 --> 00:44:20.375
have right now and what the future might look like, then DX Analytics
466
00:44:20.835 --> 00:44:26.740
might really help in providing a little bit more insight than the typical approaches allow you. And this is, I guess,
467
00:44:27.280 --> 00:44:32.410
mainly true because we have implemented something which is very computationally demanding in a sense,
468
00:44:33.465 --> 00:44:37.005
that you need lots of compute power, lots of memory, actually.
469
00:44:37.305 --> 00:44:45.349
But this was actually 1 of the or these are 2 hypotheses that we started with that that we said. Well, in these days when we start implementing a new library,
470
00:44:46.210 --> 00:44:46.789
we shouldn't
471
00:44:47.490 --> 00:44:52.365
care too much about restriction when it comes to compute power. We today have easy scalability.
472
00:44:52.665 --> 00:44:54.685
We have easy means to parallelize
473
00:44:55.785 --> 00:45:09.575
code execution even with Python. This is easily accomplished. So we started out saying no matter how computation defining an approach might be, if it's a better approach, let's go for it. So these are the major premises if you like that we say, well, we work
474
00:45:10.035 --> 00:45:11.494
at least with the working hypothesis
475
00:45:11.875 --> 00:45:12.695
that we say,
476
00:45:13.075 --> 00:45:14.215
compute and memory,
477
00:45:14.675 --> 00:45:16.055
actually are unlimited,
478
00:45:16.420 --> 00:45:17.080
are available,
479
00:45:17.380 --> 00:45:18.200
or at least
480
00:45:18.540 --> 00:45:19.040
all
481
00:45:19.380 --> 00:45:25.320
that all that we need is affordable these days. This is in the end, what we see there. 10 or 20 years ago,
482
00:45:29.494 --> 00:45:37.420
is available with a few clicks, actually, and a few dollars, actually, hasn't been available. You might have thought differently, but we started out doing this,
483
00:45:37.900 --> 00:45:54.359
on the premise that we said will compute and and where we are not binding for what we for what we do. So let's go for the best numerical methods and and even if there might be a little bit more complex. So in that sense, you have to pay a price for the additional insights, but I think you mentioned the crisis.
484
00:45:54.819 --> 00:45:56.599
When you can avoid the next crisis,
485
00:45:57.025 --> 00:46:01.845
I wouldn't go as far as that. But when you can avoid personal crisis or portfolio crisis,
486
00:46:02.145 --> 00:46:09.910
I think this is, when it pays when it pays off that you have to invest a little bit more maybe in the compute power, that you need to do that.
487
00:46:11.895 --> 00:46:17.035
So could you please walk us through how a simple analysis using DX analytics
might work?
488
00:46:18.295 --> 00:46:18.795
Actually,
489
00:46:20.040 --> 00:46:27.260
I mentioned before, this is all based on global, valuation approach. It's kind of a technical term, which in the end means that all the risk factors
490
00:46:27.935 --> 00:46:30.835
that you need to model your portfolio are
491
00:46:31.135 --> 00:46:32.435
modeled non redundantly.
492
00:46:33.215 --> 00:46:36.275
For example, again, think of the S and P 500, for example,
493
00:46:36.780 --> 00:46:37.680
You might start
494
00:46:37.980 --> 00:46:42.400
modeling the s and p 500 with a model which you think is kind of a good choice.
495
00:46:42.860 --> 00:46:47.025
The most simple 1 might be a geometric boundary motion, which dates back to 1973
496
00:46:47.724 --> 00:46:53.425
where Plexigold and Merton published their path breaking, research about the valuation of European options.
497
00:46:53.950 --> 00:46:57.170
But we also provide in DX100, it's much more sophisticated
498
00:46:57.470 --> 00:46:59.490
models involving stochastic volatility
499
00:46:59.950 --> 00:47:00.670
or jump,
500
00:47:01.069 --> 00:47:09.305
diffusion parts and so forth. So you have a great choice of different models that you can use. Once you have modeled your risk factors, again, think of the S and P 500
501
00:47:09.924 --> 00:47:13.870
or the Euro 50, for example, being, me here in Europe.
502
00:47:14.410 --> 00:47:16.650
Then you go on and model the,
503
00:47:17.050 --> 00:47:28.240
the relative positions that you have. For example, think of, European call option on the s and p 5 100 and maybe an American option on the Eurostox 50. So then you have your 2,
504
00:47:29.260 --> 00:47:31.680
different options based on 2 different risk factors,
505
00:47:32.540 --> 00:47:38.695
and, the next step would then be to combine that to a derivatives portfolio, how it is called in the external analytics.
506
00:47:39.075 --> 00:47:49.970
You have, for example, to define whether the risk factors are correlated here in this case, we would guess yes. The S and P 500 and Euro stocks, they might not be perfectly correlated, but there's
507
00:47:50.750 --> 00:47:53.010
probably some correlations, some positive correlation.
508
00:47:53.555 --> 00:48:02.700
You model that, you provide all the information, the market environment information to the portfolio class, and portfolio class in the end then takes care of the valuation,
509
00:48:03.240 --> 00:48:04.700
taking care of all the interdependencies
510
00:48:05.080 --> 00:48:05.980
of the correlations.
511
00:48:06.520 --> 00:48:08.780
The random numbers are generated. The simulate
512
00:48:10.195 --> 00:48:22.620
are are carried out, and they are carried out in kind of consistent fashion that you say, well, if we do 5 500, 000 simulations for the Euro stocks and the S and p 500, then you can be sure in the end that they are done in a coherent and consistent
513
00:48:22.920 --> 00:48:25.820
manner that you can add up values path
by path,
514
00:48:26.280 --> 00:48:26.780
scenario
515
00:48:27.515 --> 00:48:28.015
scenario
516
00:48:28.475 --> 00:48:30.255
that you have in the end risk,
517
00:48:30.635 --> 00:48:34.575
statistics that are consistent in a sense that everything has been modeled consistently.
518
00:48:34.955 --> 00:48:36.975
Everything has been simulated consistently,
519
00:48:37.420 --> 00:48:48.035
And you end up with something where I say, well, we are here, in a very consistent world and we can trust the numbers at least when it comes to the aggregation. And this is what I always had in mind and what global valuation,
520
00:48:48.655 --> 00:48:53.315
is in the end all about. And I and I stress that point because sometimes you you see,
521
00:48:53.775 --> 00:48:58.750
derivatives libraries used in the market where they use for example, 1 numerical method like a binomial
522
00:48:59.130 --> 00:48:59.950
or trinomial,
523
00:49:00.650 --> 00:49:03.515
a lattice approach for the 1 option, then they use
524
00:49:04.315 --> 00:49:17.430
a multicolor simulation for a more complex 1, and then they use a finite difference scheme for the third 1. And I say, well, the first has a value of 5, the next of 10, and the and the third 1 of 20, and then they add up the numbers. But,
525
00:49:18.210 --> 00:49:19.829
I'm always a little bit,
526
00:49:20.425 --> 00:49:23.325
yeah, well, I think it's a little bit, suspicious,
527
00:49:24.025 --> 00:49:29.870
that they say, well, I have 3 different methods. They spit out numbers, and in the end, I simply add these numbers up.
528
00:49:30.350 --> 00:49:40.485
So what we were always looking for is kind of the unified approach, which again, cost a little bit more compute power, but, we then end up with consistent numbers when it comes to values and risk.
529
00:49:41.105 --> 00:49:41.585
So,
530
00:49:42.065 --> 00:49:46.030
and, again, we have the software and the hardware available, which allows us to do things,
531
00:49:46.670 --> 00:49:49.650
which were impossible 10 or 15 years ago.
532
00:49:51.630 --> 00:49:52.610
And continuing,
533
00:49:53.230 --> 00:50:02.370
this conversation about some of the work you've done with, your company, the Python Quants, what other kinds of services do you provide and what kinds of organizations typically hire you for that?
534
00:50:03.650 --> 00:50:06.230
Well, again, it's all centered around Python for finance.
535
00:50:06.930 --> 00:50:11.270
So we I would say we cover more or less the complete value chain in this regard.
536
00:50:12.165 --> 00:50:18.185
So we provide technology, I mentioned DX analytics before, but we have also an offering where we provide
537
00:50:18.565 --> 00:50:19.625
a an interactive,
538
00:50:21.089 --> 00:50:23.349
yeah, data science, financial analytics platform,
539
00:50:24.210 --> 00:50:25.349
which is called datapark.io,
540
00:50:27.170 --> 00:50:33.235
which is the link, people might wanna have a look at. You can sign up. There are free
trials, and usually for the,
541
00:50:34.035 --> 00:50:36.375
for the for the personal individual accounts,
542
00:50:36.835 --> 00:50:39.335
we don't charge anything, and they remain open,
543
00:50:39.955 --> 00:50:41.880
at at least, for the moment.
544
00:50:42.980 --> 00:50:49.079
This is in a professional context, we call that a quant platform. I think the quant platform might be a little more expressive
545
00:50:49.454 --> 00:50:51.395
in a sense that we always imagine
546
00:50:51.775 --> 00:50:54.515
a platform where quants stop working,
547
00:50:54.895 --> 00:50:56.275
doing their stuff, implementing
548
00:50:56.575 --> 00:50:59.270
things, and in the end, not only doing interactive,
549
00:50:59.970 --> 00:51:01.750
analytics, for example, but also
550
00:51:02.050 --> 00:51:03.270
deploy, for example,
551
00:51:03.730 --> 00:51:10.515
a Python based application. But not only Python. I mean, the the, the platform itself is more or less, language agnostic.
552
00:51:10.975 --> 00:51:14.195
It's centered around the Jupyter Notebook, which is, I guess,
553
00:51:14.560 --> 00:51:17.540
the most amazing tool that our ecosystem has to offer,
554
00:51:17.920 --> 00:51:19.200
but recently they,
555
00:51:20.000 --> 00:51:24.020
they made it completely language agnostic. So we, for example, also provide Julia
556
00:51:24.355 --> 00:51:25.895
on the platform, we provide
557
00:51:26.195 --> 00:51:29.575
R as a language on the platform and much much more.
558
00:51:30.115 --> 00:51:30.855
So technology
559
00:51:32.310 --> 00:51:40.035
is 1 part of what we do at the Python Quans. Again, the Quans platform or Datapark IO is our infrastructure where people can work on. Analytics
560
00:51:40.575 --> 00:51:44.915
is kind of the financial library that people can use to attack their more sophisticated
561
00:51:45.295 --> 00:51:47.315
derivatives and risk analytics problems,
562
00:51:48.250 --> 00:51:52.510
and a few more minor things that we provide, but these are the 2 major pillars, actually.
563
00:51:53.050 --> 00:51:54.330
Then around that,
564
00:51:54.810 --> 00:51:55.950
I have also written
565
00:51:56.595 --> 00:52:02.775
a few books. My 1 book from O'Reilly is called Python for Finance. It was published last year just before Christmas,
566
00:52:03.714 --> 00:52:04.214
2014.
567
00:52:05.160 --> 00:52:07.180
Actually, it's it's selling quite well
568
00:52:07.480 --> 00:52:11.420
all across, the world. So I'm really happy about that. I think it's also an indicator,
569
00:52:12.245 --> 00:52:20.265
for the popularity of Python and finance these days. Then this July, my second book came out. It's called Derivatives Analytics with Python.
570
00:52:20.700 --> 00:52:22.480
And actually taking the 2 books together
571
00:52:22.860 --> 00:52:25.520
is actually a very good starting point and more or less,
572
00:52:25.980 --> 00:52:30.615
all what DX Analytics is about is explained in the combination of these 2 books. So,
573
00:52:30.994 --> 00:52:33.895
we have, of course, a website. I mentioned before the x dash analytics.com.
574
00:52:34.434 --> 00:52:43.100
But if you want to understand what's going on behind the scenes of DX Analytics, this is explained in the books to realtors analytics with Python and Python for finance actually.
575
00:52:43.640 --> 00:52:50.275
And Python for finance also the book that I use when I teach Python classes, be it public ones or corporate ones,
576
00:52:50.734 --> 00:52:54.674
and this brings me to the next part of what we do. We do trainings,
577
00:52:56.050 --> 00:52:58.550
And I mentioned before, they are public ones.
578
00:52:58.850 --> 00:53:02.550
For example, during our Python Quants conference series and also
579
00:53:02.850 --> 00:53:04.605
corporate customized ones,
580
00:53:05.005 --> 00:53:07.505
that we do on a regular basis for our clients.
581
00:53:08.685 --> 00:53:09.185
And,
582
00:53:10.445 --> 00:53:10.945
actually,
583
00:53:11.645 --> 00:53:14.500
around that or typically in the context of these
584
00:53:15.060 --> 00:53:21.320
workshops, and we call them workshop in London and boot camps in New York. We have also offer Python Quants conference.
585
00:53:21.620 --> 00:53:26.385
This is actually, next thing that we do. We are we're doing our own events,
586
00:53:26.845 --> 00:53:28.225
and this is already,
587
00:53:28.605 --> 00:53:29.245
we have,
588
00:53:29.645 --> 00:53:34.450
done already 4 for Python Quantes Conferences. Next year, we will again do at least 2.
589
00:53:34.990 --> 00:53:41.694
The next 1 most probably be, at the end of April in New York again and then at the end of the year again in London.
590
00:53:42.315 --> 00:53:50.090
And we are thinking of doing it in Asia as well, but, this has not yet fixed. But at least there are plans, and I hope that will, that it will materialize
591
00:53:50.470 --> 00:53:52.010
this next year, 2016,
592
00:53:52.310 --> 00:53:55.850
that we will see the for Python Q1's conference series in
593
00:53:56.595 --> 00:53:56.915
in,
594
00:53:57.715 --> 00:54:06.750
Asia as well then. In addition to that, I've also done a conference in Germany, which is called Source for Quant Finance in cooperation with, Deutsche Borse.
595
00:54:07.370 --> 00:54:12.605
And usually, what we do with these conference is that we do this in cooperation with bigger companies, which have
596
00:54:13.244 --> 00:54:14.385
venues and infrastructure.
597
00:54:14.685 --> 00:54:20.705
Before Python 1 series, which I mentioned before, is done in cooperation with the Fitch Learning and the CQF Institute.
598
00:54:21.690 --> 00:54:25.290
And this brings me to the next kind of event that we do. This is,
599
00:54:26.130 --> 00:54:26.630
our,
600
00:54:27.369 --> 00:54:29.070
also the biggest meetup I'm running
601
00:54:29.445 --> 00:54:32.185
is in London, which is called Python for 1 Finance.
602
00:54:32.725 --> 00:54:34.505
We have now more than 1100
603
00:54:34.805 -->
00:54:38.250
members in this group, and it's quite an active group. We have,
604
00:54:38.869 --> 00:54:43.930
7 to 8 bigger meetups with around 100 people per year in London,
605
00:54:44.470 --> 00:54:53.275
and it's again quite an active group and it always send the raw Python for confinance with typically speakers from very diverse fields in our industry
606
00:54:53.690 --> 00:54:58.190
It's always very, very useful, I guess, for all participants. So if you're located in London,
607
00:54:58.490 --> 00:55:03.515
give it a try, and I'm pretty sure at the end of January or beginning of February, we'll have our next
608
00:55:04.075 --> 00:55:08.974
1 there again. And this is supported usually by, Thomson Reuters, which is the partner there.
609
00:55:09.355 --> 00:55:13.135
So we are pretty happy that we have these strong big partners. We support
610
00:55:13.640 --> 00:55:15.900
our purpose and and which makes it
611
00:55:16.280 --> 00:55:18.060
possible that we can do these
612
00:55:18.440 --> 00:55:19.260
amazing events,
613
00:55:19.800 --> 00:55:21.100
around the globe actually.
614
00:55:21.720 --> 00:55:25.395
And what we live off, actually, this is the last part,
615
00:55:25.715 --> 00:55:31.175
but not least more, is consulting and development work. So, when it comes to revenues
616
00:55:31.530 --> 00:55:32.350
and income streams,
617
00:55:32.890 --> 00:55:38.350
the consulting and the and the development part, by far the the most important ones that we consult,
618
00:55:38.724 --> 00:55:42.744
a bigger financial institutions or asset managers or,
619
00:55:43.525 --> 00:55:44.025
exchanges,
620
00:55:45.444 --> 00:55:49.140
hedge funds and so forth when it comes to Python for Quant Finance.
621
00:55:49.920 --> 00:55:56.420
And there, it is very, very different what we do there. For example, I mentioned before, applying Deeks Analytics in practice,
622
00:55:56.825 --> 00:56:00.765
running production code, that we've implemented forward with institution risk
623
00:56:01.145 --> 00:56:08.490
based models based on Python and so forth. So it's quite diverse, but again, in the end, it's all about Python for finance and in particular,
624
00:56:08.869 --> 00:56:10.250
the computational finance,
625
00:56:10.710 --> 00:56:13.765
part and the financial data science part is what we focus on.
626
00:56:14.325 --> 00:56:23.305
And, yeah, you see, we are only small company, but this is what we do on a daily basis. And I think we cover almost all the bases when it comes to this
627
00:56:23.830 --> 00:56:24.730
niche of the
628
00:56:25.190 --> 00:56:25.690
market.
629
00:56:26.150 --> 00:56:28.490
It's very cool that you folks have managed
630
00:56:29.350 --> 00:56:39.565
to keep the, you know, keep the lights on and and get enough consulting income and yet have that work allow you to contribute back to open source? Because I know sometimes big enterprises
631
00:56:40.330 --> 00:56:42.190
not exactly open source friendly.
632
00:56:43.050 --> 00:56:45.010
Yeah. That's that's true. That's true. I mean,
633
00:56:45.450 --> 00:56:46.330
in the end, but,
634
00:56:47.130 --> 00:56:49.085
yeah, I've always this discussion with
635
00:56:49.645 --> 00:56:52.705
living out of open source. I think you should contribute back,
636
00:56:53.085 --> 00:56:54.625
to open source, and,
637
00:56:55.724 --> 00:57:02.420
it's also kind of a model that you can choose these days, and I guess the the whole culture and and the infrastructure
638
00:57:03.039 --> 00:57:07.964
have changed kind kind of a bit that you can say, well, you can't even live off your open source work.
639
00:57:08.665 --> 00:57:16.980
And this is kind of fascinating. They say, well, you share with the world, people contribute back, and you can still make a living out of something that you share, which might be counterintuitive
640
00:57:17.280 --> 00:57:18.020
for people,
641
00:57:18.800 --> 00:57:34.390
yeah, who are not used to these things. But more and more, and I've seen seen, and I've joined that, joined it over the year. The people saying, well, if what you did, for example, inspired me to open source, for example, the the 1 example that I mentioned from Sayid Amin, from the Thelesians.
642
00:57:34.690 --> 00:57:41.484
And he said, well, I've seen so many, good talks about people who argued why it is good to open source libraries and this actually,
643
00:57:42.345 --> 00:58:01.235
helped me made up my mind and and make the decision to open source Python lessons as well. So this is actually something that I really enjoy that people pick up, the idea and the culture and say, well, we go that way. But on the other hand, we still have proprietary technology. For example, the the the platform part and what we provide there is not open source yet, I would say.
644
00:58:01.935 --> 00:58:14.195
I didn't see the point to do that. But again, here we provided for free more or less. So we have many, many students working on the platform. We have a few 1, 000 users on the platform, and I know that many students and so forth that might not have to budget,
645
00:58:14.495 --> 00:58:30.020
to afford kind of professional environment like that. They're really happy that they can use that for free, this nice service integrated platform. So we try our best to provide, good value here back to the community as well even if this particular thing is not, open sourced yet again.
646
00:58:32.605 --> 00:58:35.905
So is there anything that we should have asked you but didn't?
647
00:58:36.525 --> 00:58:39.905
Or are there any places where our listeners can help
648
00:58:40.210 --> 00:58:42.790
contribute to the open source work that you folks do?
649
00:58:43.330 --> 00:58:47.190
Yeah. For sure. I mean, we we are running a few websites. So it's it's difficult
650
00:58:47.810 --> 00:59:02.950
to point people to a single page. But a good starting point for people who are interested in what we do, and there you find almost
all the links to the other stuff, like to the books, to the media group, to the events pages, and so forth is our company page, which is, h t p,
651
00:59:04.210 --> 00:59:05.910
colon slash slash, tpqdot
652
00:59:06.335 --> 00:59:07.875
io. Again, tpq.io
653
00:59:08.414 --> 00:59:09.714
for the Python quants.
654
00:59:11.214 --> 00:59:16.035
There you find, again, links to many, many other places and resources. You find the link, for example,
655
00:59:16.530 --> 00:59:19.109
again, to DX Analytics. We have, of course,
656
00:59:19.570 --> 00:59:20.790
our GitHub repo,
657
00:59:21.570 --> 00:59:26.535
up with, the code of DX Analytics. You can try it out on on, the quant platform.
658
00:59:27.154 --> 00:59:28.454
The link for that is quantashplatform.com.
659
00:59:29.714 --> 00:59:31.494
If you prefer more the datapark,
660
00:59:33.075 --> 00:59:34.134
styling. Actually,
661
00:59:34.670 --> 00:59:39.250
the technology behind it is the same, but the styling is a little bit different. You can use Datapark dot io.
662
00:59:39.950 --> 00:59:41.410
It's more or less the same.
663
00:59:41.870 --> 00:59:45.465
So go to these places, have a look, and if you have questions, yeah, just,
664
00:59:45.865 --> 00:59:48.365
either drop me an email. You'll find their contact details,
665
00:59:48.745 --> 00:59:52.125
or you can also follow me on Twitter. My Twitter handle is,
666
00:59:52.585 --> 00:59:58.310
d y g h or just Google me up. You should find lots of, information about myself.
667
01:00:00.290 --> 01:00:03.265
Yeah. I see you're you're writing it down here, actually.
668
01:00:03.565 --> 01:00:13.569
But Twitter is kind of a good way to stay in touch and to see what we are doing. You get usually the latest news with regard to conference we're doing or if there's an upcoming meetup, and so forth.
669
01:00:14.109 --> 01:00:15.970
So it's kind of a good way to keep up,
670
01:00:16.829 --> 01:00:19.010
and to get it in a more interactive fashion.
671
01:00:19.550 --> 01:00:20.770
And, again, go to tpq.il,
672
01:00:21.309 --> 01:00:28.345
and then you'll find many, many links to many, many other resources that you can use, also to my private website, hillippish.com,
673
01:00:29.125 --> 01:00:31.224
where I usually store all the presentations
674
01:00:31.525 --> 01:00:32.190
that I give,
675
01:00:32.750 --> 01:00:35.970
where you also find the, videos and so forth.
676
01:00:37.310 --> 01:00:37.810
And,
677
01:00:38.977 --> 01:00:43.705
yeah, I think there's a lot there. And people who are interested, just just get in touch. I'm happy
678
01:00:44.085 --> 01:00:44.825
to discuss,
679
01:00:45.205 --> 01:00:45.705
things.
680
01:00:46.005 --> 01:00:49.609
So I see, again, my, my Twitter handle is j like
681
01:00:50.630 --> 01:00:51.130
Johannes.
682
01:00:51.510 --> 01:00:53.530
It's d y. Not g.
683
01:00:54.630 --> 01:00:56.089
It's j h.
684
01:00:57.855 --> 01:00:58.355
Well,
685
01:00:58.655 --> 01:01:00.435
thank you. This has been a
lot of fun.
686
01:01:01.215 --> 01:01:05.635
So if there is nothing else before we move to the pics, then we can go ahead and do that.
687
01:01:05.970 --> 01:01:07.670
Alright. So I will get us started.
688
01:01:08.130 --> 01:01:14.789
So this week I wanna pick a novel that I've been reading. I'm about halfway through it. It's Kraken by China Meeval.
689
01:01:15.625 --> 01:01:19.005
I mentioned him as an author before on 1 of the other episodes,
690
01:01:19.305 --> 01:01:19.805
and
691
01:01:20.665 --> 01:01:22.605
this is in his typical fashion
692
01:01:23.280 --> 01:01:32.565
very different than any of his other novels but at the same time the character of the text is the same. He does a very good job of using very innovative
693
01:01:33.065 --> 01:01:33.565
and
694
01:01:34.065 --> 01:01:35.125
evocative language.
695
01:01:35.905 --> 01:01:36.965
And it's a story,
696
01:01:37.505 --> 01:01:47.670
it starts off it seems like it's just going to be a regular kind of mystery story but it very quickly incorporates a lot of supernatural elements that make it very
697
01:01:48.455 --> 01:01:48.955
compelling.
698
01:01:49.495 --> 01:01:53.435
And it does a really good job of juxtaposing those elements against the normal
699
01:01:53.815 --> 01:01:57.435
everyday world that the story is taking place in. It's set in London,
700
01:01:58.000 --> 01:02:00.420
Definitely a great novel. I recommend checking it out.
701
01:02:01.680 --> 01:02:07.140
My next pick is a series of books that I've been reading with my son called Heroes in Training,
702
01:02:07.755 --> 01:02:11.935
and there are I think 7 or 8 books in total and they're all about the
703
01:02:12.474 --> 01:02:13.455
gods of Olympus
704
01:02:13.915 --> 01:02:17.055
at the age of about 10 and it's the story of them
705
01:02:17.609 --> 01:02:18.509
trying to
706
01:02:18.970 --> 01:02:20.269
overthrow Kronos.
707
01:02:20.890 --> 01:02:24.190
And so it's got a lot of good humor baked into it,
708
01:02:24.890 --> 01:02:29.905
very well targeted at a younger age range, so been having a lot of fun reading that with him.
709
01:02:31.165 --> 01:02:38.289
Along the same lines, there's a set of graphic novels that I've been, that I got for him and I've been reading as well called Olympians.
710
01:02:39.470 --> 01:02:48.444
And the I believe the author's name is George O'Connor. And it's just very well done, very well researched graphic novels that present the gods of Olympus
711
01:02:49.385 --> 01:02:58.310
and you know I've done a fair bit of reading about different Greek mythology and there were still a lot of elements that showed up in those graphic novels that I had never come across before. So
712
01:02:58.705 --> 01:03:01.525
definitely interesting, definitely recommend reading those as well.
713
01:03:01.905 --> 01:03:08.630
And my last pick, kind of along the lines of our data focus here is the Data Elixir newsletter
714
01:03:09.010 --> 01:03:17.575
and it's just a weekly newsletter that has a bunch of links to interesting and useful articles about what's happening and the different aspects of data and data science.
715
01:03:18.515 --> 01:03:20.535
And with that, I will pass it to you, Chris.
716
01:03:23.315 --> 01:03:23.750
So,
717
01:03:24.790 --> 01:03:35.944
I was just in Vermont for a week. We typically go there every Christmas. And for me, Vermont is many things, beautiful scenery, great food, and also, it is home of really, really great beer.
718
01:03:36.484 --> 01:03:41.790
So I have a few beer picks this time. My first pick is from a Vermont brewery called Hill Farmstead
719
01:03:42.490 --> 01:03:43.210
Brewing, and,
720
01:03:43.610 --> 01:03:49.745
the beer is called Edward because, apparently, that's their grandfather was named Edward. They made this ale in his honor.
721
01:03:50.305 --> 01:03:52.485
It's an American pale ale.
722
01:03:53.105 --> 01:03:54.005
It's really
723
01:03:54.305 --> 01:03:59.285
interesting. It's it's got a really incredible aroma, and it is
724
01:03:59.790 --> 01:04:00.530
really kind
725
01:04:00.910 --> 01:04:02.210
of hopsy and tart,
726
01:04:03.310 --> 01:04:12.915
with flowery and pine notes, which I know sounds kinda strange drinking a beer with flower and pine, but it's actually really delicious and and and definitely worth hunting down.
727
01:04:13.535 --> 01:04:14.915
My next pick is,
728
01:04:15.615 --> 01:04:26.665
long trail brewing. A lot of people who live in New England kinda know Long Trail because of what they find from them in bottles and stores. And that's you know, those beers are good. But if you ever manage to get to,
729
01:04:27.065 --> 01:04:28.125
Bridgewater Corners,
730
01:04:28.744 --> 01:04:36.089
where the actual plant is, they have a number of special beers that you can only get there on tap that are fantastic.
731
01:04:36.470 --> 01:04:38.010
The first 1 is called culmination.
732
01:04:38.549 --> 01:04:43.530
It's a chocolate porter. I don't usually like chocolate porters, but this 1 is really
733
01:04:44.105 --> 01:04:45.245
exceptional. It's
734
01:04:45.625 --> 01:04:48.045
less sweet than the average. It's,
735
01:04:48.745 --> 01:04:50.045
just got a lot of complexity
736
01:04:50.345 --> 01:04:58.130
and depth. It's it's really good stuff. The next 1 is gonna sound really cheeky because of the name, but it's a really excellent beer. It's called,
737
01:04:58.530 --> 01:05:00.230
space juice, double IPA.
738
01:05:01.555 --> 01:05:02.295
It is
739
01:05:02.675 --> 01:05:04.775
a seriously, seriously hoppy,
740
01:05:06.115 --> 01:05:06.615
IPA,
741
01:05:07.910 --> 01:05:08.410
caramel
742
01:05:09.069 --> 01:05:09.569
malts,
743
01:05:10.309 --> 01:05:11.210
really smooth,
744
01:05:11.829 --> 01:05:16.170
and I because I was, again, I was there at the at the at the actual brewpub.
745
01:05:16.965 --> 01:05:20.905
They happen to have it cast conditioned on tap, which was just delightful.
746
01:05:22.245 --> 01:05:32.920
So that's definitely worth checking out. My last pick is amazing amazing, a Python pick. I usually don't have much to pick because usually there are smarter people on here than I
747
01:05:33.244 --> 01:05:40.944
discussing, you know, Python things that that are much more interesting. But I just encountered something that really hit a sweet spot for me. It's called Flask
748
01:05:41.565 --> 01:05:42.464
dash restless,
749
01:05:43.089 --> 01:05:48.230
And essentially, it allows you to take your sort of plain old Python classes,
750
01:05:49.569 --> 01:05:51.990
and make them into a REST API.
751
01:05:52.625 --> 01:05:59.365
And what I really like about it is that in addition to being sort of very easy to set up and hook up, it interfaces with SQLAlchemy
752
01:06:01.520 --> 01:06:02.340
very nicely.
753
01:06:03.200 --> 01:06:07.860
What you end up with is a Flask application, and so
754
01:06:08.275 --> 01:06:11.815
all of these sort of, you know, other pieces of the Flask ecosystem,
755
01:06:12.675 --> 01:06:13.415
like authentication
756
01:06:14.195 --> 01:06:17.255
or whatever other kind of plug ins you may wanna use,
757
01:06:18.019 --> 01:06:19.240
will probably work.
758
01:06:20.019 --> 01:06:30.315
So I've really been enjoying this. I got a side project going that I wanna write a web service for. And of all of the things of of this ilk out there, this 1 just fit the best for me.
759
01:06:30.855 --> 01:06:37.890
That's it for me this week. Yves, what do you have for us for picks? Well, I've also a few books actually. Currently, I think it's a good time to read.
760
01:06:38.349 --> 01:06:41.695
Usually, when there's kind of the middle of the year and you're busy busy working.
761
01:06:42.095 --> 01:06:45.935
But a fascinating book that I read recently is kinda it's called The Willpower Instinct.
762
01:06:46.335 --> 01:06:50.650
I think it's around since quite a while. I think it was published 2011 by Kelly McGonigal.
763
01:06:51.030 --> 01:07:01.675
It's kind of a fascinating book about, how how humans function actually and how you might not be able to change your behavior, how how you get caught in your habits, and what's going on
764
01:07:01.975 --> 01:07:02.475
chemically,
765
01:07:02.775 --> 01:07:06.609
psychologically. And I I really wanna I I listen to the book actually,
766
01:07:07.150 --> 01:07:09.570
my my iPhone, but I also read it
767
01:07:09.950 --> 01:07:13.970
on paper. But it was really, like, kind of every chapter was, like,
768
01:07:14.385 --> 01:07:23.125
I had the feeling at least about myself and and how I miss recently doing this or that and so forth. So if you're interested in changing your behavior and and or at least to understand,
769
01:07:23.740 --> 01:07:25.680
how man how human mankind,
770
01:07:26.060 --> 01:07:31.040
actually is working, this is kind of a very good book. The best, at least in this regard, I know actually.
771
01:07:31.615 --> 01:07:40.210
Another book, I am reading currently
is The Way of the Seal. It's kind of a nice book by Mark Divine. If you wanna get stronger and wanna also kind of develop willpower,
772
01:07:40.670 --> 01:07:44.290
in the way the seals do, and I guess they're 1 of the best, elites,
773
01:07:45.070 --> 01:07:47.645
units in the world to actually special forces.
774
01:07:48.585 --> 01:07:55.380
So it's kind of also a good book, a little bit more practical 1 and that there are many, many exercises how you can challenge yourself,
775
01:07:56.240 --> 01:08:00.079
to get maybe a better person. Another 1 a little bit different, is called,
776
01:08:00.640 --> 01:08:01.140
Sapiens,
777
01:08:01.760 --> 01:08:06.575
a previous 3 volume of mankind, but you will know a Harari, a kind of a nice book,
778
01:08:07.055 --> 01:08:08.355
from the very beginning,
779
01:08:08.655 --> 01:08:10.675
maybe 20, 30000 years ago,
780
01:08:11.030 --> 01:08:16.890
Homer Erectus, was around and then Neanderthals and so forth, up to today. So the scientific revolution
781
01:08:17.430 --> 01:08:25.335
actually is a very nice read. Usually, nothing I read, but again, during the holidays, kind of very nice read for a little bit, better understanding,
782
01:08:25.955 --> 01:08:27.175
what shaped actually,
783
01:08:28.210 --> 01:08:32.470
our, yeah, our species, if you like, like religion and agricultural,
784
01:08:33.410 --> 01:08:34.870
and evolution and so forth.
785
01:08:35.495 --> 01:08:41.435
Also very, very, yeah, short while read, actually, very well written. And the last 1 is also a Python 1,
786
01:08:42.150 --> 01:08:46.730
It's also had the Python pick. I wanna, recommend the book high performance computing
787
01:08:47.510 --> 01:08:49.929
Python by Ian Oswald and and Michelle,
788
01:08:50.630 --> 01:08:51.130
Gralik.
789
01:08:51.775 --> 01:08:53.554
It's kind of a very good
790
01:08:53.935 --> 01:08:57.474
book from my point of view, which is not only about high performance,
791
01:08:58.335 --> 01:09:01.280
Python actually, but it's more or less about,
792
01:09:01.659 --> 01:09:04.079
yeah, how Python, how the interpreter works,
793
01:09:04.380 --> 01:09:09.595
and and what is going on behind the scenes. So we discussed many, many things today in this podcast,
794
01:09:10.315 --> 01:09:11.615
how to make Python,
795
01:09:12.155 --> 01:09:34.190
more fast and so forth. But the very reasons and and the workings behind the scenes when it comes to data structures, data storage, and and then any critique kind of, details that you have to be, aware of if you wanna go professional in this regard is explained very well from my point of view in this book. So I really like it. And, if you wanna do a Python for Quant Finance, this might be 1 of the books that should be on your on your bookshelf actually.
796
01:09:36.330 --> 01:09:37.230
Very interesting.
797
01:09:37.850 --> 01:09:44.785
Well, thank you again for taking the time out of your day to join us and tell us all some more about your work in quantitative finance.
798
01:09:46.460 --> 01:09:53.600
You've already told us about how to get in touch with you. So, yeah, thank you very much, and I hope you enjoy the rest of your evening as well as your new year.
799
01:09:54.165 --> 01:09:59.705
Thank you, Chris. Tobias, thank you for having me, and, yeah, happy new year, and all the best for 2016.
800
01:10:00.165 --> 01:10:02.025
Happy new year to you as well. Bye bye.