Forums  > Books & Papers  > Opinions and recommendation on books  
     
Page 1 of 1
Display using:  

Maggette


Total Posts: 1301
Joined: Jun 2007
 
Posted: 2021-01-10 15:06
I thought it would be nice to have a thread where you can ask for opinions on a book before buying it.

I will make the start:

I know we haven expert on Random Matrix Theory on this phorum.


A First Course In Random Matrix Theory

By Potter and Bouchaud. So far I liked stuff from Bouchaud. Opinions (by our RM expert or anybody else)?

Ich kam hierher und sah dich und deine Leute lächeln, und sagte mir: Maggette, scheiss auf den small talk, lass lieber deine Fäuste sprechen...

dgn2


Total Posts: 2077
Joined: May 2004
 
Posted: 2021-01-11 21:10
I would be really interested in an informed opinion about this book as well.

...WARNING: I am an optimal f'er

bandi_np


Total Posts: 10
Joined: Nov 2020
 
Posted: 2021-01-11 22:16
Some chapters of the book seem to be discussed in one of his papers https://arxiv.org/pdf/1610.08104.pdf, might be an interesting preview

Maggette


Total Posts: 1301
Joined: Jun 2007
 
Posted: 2021-01-12 10:03
Thx bandi_np. Really appreciated. I was aware that Bouchaud and Co have tons of papers RMT, but I was to lazy to work through all of them and puzzle stuff together.

But that one already looks like a short book. Thx.

Ich kam hierher und sah dich und deine Leute lächeln, und sagte mir: Maggette, scheiss auf den small talk, lass lieber deine Fäuste sprechen...

jslade


Total Posts: 1243
Joined: Feb 2007
 
Posted: 2021-01-12 16:06
I downloaded and glanced at it (you guys know about libgenesis right? not that I would EVER USE such a SINFUL thing); looks like pretty typical Bouchaud and Potters book, with lots of interesting shiny things which appeal to former physics people and not so many real world examples demonstrating why it's useful to practitioners. They posit various formal connections between RMT and, for example, shrinkage estimators, which I suppose are probably true, but aren't very informative. I already know how to use shrinkage estimators. I also don't think it's particularly insightful using a rotationally invariant estimator for out of sample estimates for the covariance matrix for reasons that should be obvious.

Anyway de gustibus. RMT is pretty cool to think about for a few minutes when you're not doing something more useful.

"Learning, n. The kind of ignorance distinguishing the studious."

nikol


Total Posts: 1345
Joined: Jun 2005
 
Posted: 2021-01-12 20:34
> not that I would EVER USE such a SINFUL thing

Innovation!
Would you buy blockchained indugence contract?
Evil Smile

... What is a man
If his chief good and market of his time
Be but to sleep and feed? (c)

chiral3
Founding Member

Total Posts: 5207
Joined: Mar 2004
 
Posted: 2021-01-12 22:40
I haven't read the book but I suspect that there isn't anything in there that you couldn't get from papers we've talked about over the past couple of decades. Ah, the zeros of zeta functions in finance, incroyable.

Nonius is Satoshi Nakamoto. 物の哀れ

nikol


Total Posts: 1345
Joined: Jun 2005
 
Posted: 2021-01-13 11:14
I havent read this book either, but scanned the article sometime ago. Nothing specific is left.

Intuitively, I believe that corr is ill formed concept coming from analysis of static samples. Conditional probability is more correct way to describe the reality. And of course, corr(a,b)'s are linked to prob(a,b)'s

Correlation causes many conceptual misinterpretations, especially related to causal structures. It is a fudge factor.

PS. Although I confess that I use corr, but still try to develop internal reasoning linked to probabilities.

... What is a man
If his chief good and market of his time
Be but to sleep and feed? (c)

dgn2


Total Posts: 2077
Joined: May 2004
 
Posted: 2021-01-14 02:51
I like the RMT stuff because it pushed me to do a bunch of MC on different cross-dependent processes and look at the distributions of the eigenvalues. Comparing the distributions of eigenvalues for actual and synthetic data makes a lot of sense to me. I would be interested in knowing why this isn't useful to others. Maybe the point is that if I have read all the main papers and done lots of exploration via MC then I don't need the book.

To me, knowing that the correlation matrix has a hierarchical structure and that there tends to be one really large eigenvalue and a few other large eigenvalues (all relative to a random matrix) is a useful stylized fact. Knowing how this changes over time is also useful. Knowing how these things look with fat-tailed distributions, shuffled returns, etc...helps me in developing systems for allocating across instruments within a strategy.

...WARNING: I am an optimal f'er

deeds


Total Posts: 512
Joined: Dec 2008
 
Posted: 2021-01-14 10:47

Hi @dgn2

agree about practitioner's utility of this technical direction...also some deep theoretical links waiting to be made with a broader set of analytical practices

hope not to digress, but would you gloss 'cross-dependent processes' in MC...are you talking about managing the correlation between the processes, looking at covariance/correlation matrices? Is there a distinction on your map between dependence and correlation?


dgn2


Total Posts: 2077
Joined: May 2004
 
Posted: 2021-01-14 12:47
To me dependence is the generic term that includes all measures not just linear correlation. With MC, you can look at the distribution of eigenvalues (or any other quantity coming out of a decomposition) for any set of processes.

I often look at things where linear correlation doesn't make sense. For instance, I might pass instrument returns to a trading system simulator and get back system returns. I then look at the linear correlation and corresponding eigenvalues for the instrument returns and the system returns and compare them. That sort of gives me a sense of the information available to the system operator versus the allocator who doesn't know how the system works in detail.

...WARNING: I am an optimal f'er

deeds


Total Posts: 512
Joined: Dec 2008
 
Posted: 2021-01-14 14:52

thank you so much for the kind response

enquiring out of interest in approaches that draw a distinction between dependence and statistical linear/nonlinear correlation or association and a particular finance problem that seems to require such a distinction (slowly shuts door of rabbit hole)

very cool application

i'm betting you've looked at it, but if not, multivariate mutual/conditional entropy (and associated ICA) may be of interest in the kinds of comparisons you describe


deeds


Total Posts: 512
Joined: Dec 2008
 
Posted: 2021-01-14 14:52
EDIT: oops, doubled

dgn2


Total Posts: 2077
Joined: May 2004
 
Posted: 2021-01-14 16:52
@Deeds: I would be interested in hearing/discussing more, but maybe let's start a separate thread or send me an email (see profile).

...WARNING: I am an optimal f'er
Previous Thread :: Next Thread 
Page 1 of 1