Forums  > General  > QuantFinance is dead  
     
Page 2 of 2Goto to page: 1, [2] Prev
Display using:  

rod


Total Posts: 392
Joined: Nov 2006
 
Posted: 2019-09-19 15:28
If mentioning PE is not threadjacking, over the past year or two I have seen a lot of attacks on PE coming from a guy named Daniel Rasmussen:

□ Daniel Rasmussen, Private Equity: overvalued and overrated?, American Affairs, Spring 2018.

□ Richard Teitelbaum, One young Harvard grad’s quixotic quest to disrupt private equity, Institutional Investor, April 25, 2018.

jslade


Total Posts: 1195
Joined: Feb 2007
 
Posted: 2019-09-19 15:38
Oh Chrikey we're not talking about this Lopez de Prado dude again are we? I have no idea who he is beyond reading his dumb-ass "machine learning in finance" book, which isn't even a good book for "machine learning for time series." Can someone explain to me why he is any more relevant than some boob who wants to sell you his really excellent forex TA system?

Googling him; he cooled his heels for 7 years at Horst Simon's LBL "financial observatory," and published dozens of goofy thought piece economics papers with titles like "the future of empirical finance," which doesn't exactly indicate a white hot career doing practical things a sane person can learn from.

Maybe he is this decade's Nassim Taleb. It looks like he sells a lot of bullshit and talks to a lot of reporters. "Crowdsourcing" quant stuff; yeah, that worked super well for Quantopian and Numerai. Lighting the world on fire!

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

rod


Total Posts: 392
Joined: Nov 2006
 
Posted: 2019-09-19 15:41
jslade, search "Lopez de Prado" on LinkedIn. You may be surprised to learn you two have LinkedIn friends in common.

riskPremium


Total Posts: 21
Joined: Nov 2018
 
Posted: 2019-09-19 15:56
@jslade

I only scanned contents of Lopez de Prado's machine learning book and was thinking about buying one. Can you elaborate why his book isn't a good one?

jslade


Total Posts: 1195
Joined: Feb 2007
 
Posted: 2019-09-19 16:24
I commented on it here:

http://nuclearphynance.com/Show%20Post.aspx?PostIDKey=186472

I encourage you to download it from a file sharing site and have a look at it first. Most of it is nonsense and filler. Ch 4 looked OK.

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

tbretagn


Total Posts: 281
Joined: Oct 2004
 
Posted: 2019-09-19 16:57
@jslade hearing someone doing some big GPU arfima lately though

Et meme si ce n'est pas vrai, il faut croire en l'histoire ancienne

riskPremium


Total Posts: 21
Joined: Nov 2018
 
Posted: 2019-09-20 02:00
@jslade

Thanks. Also very informative thread.

Maggette


Total Posts: 1167
Joined: Jun 2007
 
Posted: 2019-09-20 08:03
Young Harvard snob making a killing by buying small caps and selling it as an alternative to overhyped PE.....?

I guess he might be right that PE is overrated or probably even in bubble state....but why is he so damn sure small caps aren't either?


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

contango_and_cash


Total Posts: 114
Joined: Sep 2015
 
Posted: 2019-09-20 12:34
Actually had a call with him. Nice enough.

He strikes me more as a well networked guy who knows folks w money or started with some family backing.

What they're doing is no more technical or quant than osam's micro Quant product.

contango_and_cash


Total Posts: 114
Joined: Sep 2015
 
Posted: 2019-09-20 14:22
@jslade have you seen kolm & ritter new paper on reinforcement learning? would be curious to know what you think.

much different approach than ldp.

https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3449401

nikol


Total Posts: 850
Joined: Jun 2005
 
Posted: 2019-09-21 09:38
Let's stop this "argumentum ad hominem"

However, at weekend I am glad to hear the old song "Long live Quant'Fin" like this

https://www.youtube.com/watch?v=sLN8lHBBKck

doomanx


Total Posts: 35
Joined: Jul 2018
 
Posted: 2019-09-25 20:48
Totally agree with slade RE: prado - just another guy selling his book to make up for his lack of P&L.

@contango_and_cash my take on RL having had it deployed in production for several years - it's good for learning strategies in a sequential decision making task where you can easily simulate trajectories of the environment. Things become a lot more difficult if the environment is hard to simulate (for example, interacting with the order book will change the trajectory when trading a reasonable clip size). To that end you either need:
a) A big historical database of state-action pairs and the reaction of your environment (E.g. In the past we placed this trade with this inventory/at this price and the book did this) OR
b) a good model of the evolution of your environment.

If you're simulating from historical data and you didn't actually take the action then you don't know how your action would have affected the future trajectory and therefore you rely on a model for the evolution (and so indirectly the reward/cost/payoff) or simply a model to estimate the cost of taking an action in a state directly. This is a whole 'nother can of worms in itself and usually the hardest part of the whole thing.

Finally will just mention the dynamic programming RL algorithms blow up in algorithmic complexity exponentially with the dimension of the state space and so one usually strays towards the on policy side of things. These algorithms tend to require a (very!) large number of episodes to converge. It's also hard to write efficient software for training as RL algorithms have a lot of sequential dependency and so are hard to parallelize. That being said for my use case it outperformed HJB-methods and whatever else we tried so the effort is not misplaced.

contango_and_cash


Total Posts: 114
Joined: Sep 2015
 
Posted: 2019-09-26 17:54
@doomanx - thanks. I've been spending my time trawling sec filings which is more brute force than interesting. Starting to poke around in RL to think about math for a bit.

I appreciate the response.

doomanx


Total Posts: 35
Joined: Jul 2018
 
Posted: 2019-09-27 11:55
@contango_and_cash no worries, I feel your pain on that one. One particularly nice feature of RL is that even if you use some model to approximate the environment once you go live you can keep training with online learning. Provided you're somewhere in the right neighborhood this lets you smooth out the model error. If you're not there are several 'exploration' strategies (idea is you perhaps randomly try an action you currently estimate to be 'worse' as the current 'best' may just be a local optima).

It's also nice for HFT as once the strategy is trained it's either a Q-table (array) mapping state to action or if the state space is big a function approximating this mapping which only requires you to keep a small number of parameters in the cache, whereas the PDE approach you need to keep solving and then pre-caching the values around your current 'state' in a grid, which generally requires a lot more work.

Rashomon


Total Posts: 208
Joined: Mar 2011
 
Posted: 2019-11-14 15:07
jslade: he got hired by numerai as head of something.

nikol: yes, ad hominem is bad, but look at what’s going on: bad research out of name-brand schools, ok research out of advertising monopolies, and said advertising firms pull from the professoriate. And the largest quant fund half-times their people at Stern + uses Chicago logic to recruit aum. And the largest fund hires an EECS prof as “head of AI”. There’s a certain decency to taking people at their word, which doesn’t fit this indecent world.

I’m always super down to discuss solid papers and books, even if the application is outside banking/trading/funds/etc. But it is worth pointing out, even if just for the lurkers, that Millennium spawned WorldQuant; that the head of Ann Arbor’s financial engineering degree has never worked anywhere in the industry; that Ann Arbor’s MFE brags about a 26-year-old immigrant “becoming a PM” (by which they literally mean she’s listed on the filing for a smart-beta ETF). People can figure this stuff out about Jim Dalton and Larry Williams on their own. When sufficiently much mathematics is used to obscure the point, they will look here.

Kitno


Total Posts: 403
Joined: Mar 2005
 
Posted: 2019-11-20 10:49
@rod I like a lot of what Dan writes. He and Greg distribute weekly newsletters via e-mail on market thoughts. I've chatted to him a number of times.

Somewhat embarrassing I assumed for the last 3y he got my PA e-mail as I worked with him before (I thought he had been a HY dealer), but it turns out after looking at LinkedIn I didn't.....! Head against Wall

"Yeh, after that blow out I bid the bonds at 76 and you hit man...You're 77/81 now? Cool man...What? Do I care at 80? No mate... I'm 73 bid now...I'm sure you didn't just load up just for me...".
Previous Thread :: Next Thread 
Page 2 of 2Goto to page: 1, [2] Prev