 jungle
|
Chief Rhythm OfficerCSD LLC |
Total Posts: 3169 |
Joined: Jul 2004 |
|
|
my latest idea.....
...is an interesting one. machines find ideas, people approve them, rather than people find ideas (qualitative hypotheses) and machines (via backtesting) approve them. |
|
|
|
 |
 Nonius
|
Founding MemberNonius Unbound |
Total Posts: 12808 |
Joined: Mar 2004 |
|
|
yes, and to clarify...the monkeys can say NO, but they cannot say, hey, doood, I was looking at Ford and I think it is oversold...bla bla...
the discipline is that you let the machines do a first pass, and then you let the monkeys approve (and possibly nix) the list one by one....
oh, but the machines decide (once the list is gelled), what the trades will be and what size. |
Chiral is Tyler Durden |
|
 |
 kr
|
Founding MemberNP Raider |
Total Posts: 3561 |
Joined: Apr 2004 |
|
|
yeah, that's just what I want to be - downstream from a fucking COMPUTER
yes, a lot of the 'fundamentals' business model is flawed, but this is no way to fix it
I am trying to understand how this 'computer intelligence' is supposed to interface with 'real intelligence', and I don't get it. But please tell all your fund managers this is a great idea, and you can even lend them billions to do it. Yeah, completely changed my mind just this minute, it's a fucking great idea. Brilliant. |
my bank got pwnd |
|
|
 |
 NIP247
|
|
Total Posts: 559 |
Joined: Feb 2005 |
|
| |
 |
 Nonius
|
Founding MemberNonius Unbound |
Total Posts: 12808 |
Joined: Mar 2004 |
|
|
for your space, it is definitely not a great idea....that is for sure...
for long/short equity, some flavors of credit, some flavors of CB, some trend following, and obviously stat arb, it is probably the right idea....
I'm down on just quant, and definitely down on just fundamental (except for strategies that require peering deeply into work out, details of the waterfall in an ABS, etc etc....) I would just like to speed up the strategy discovery process......and, I'd like to do it on a large scale basis so that one could find dozens, if not hundreds, of viable unrelated strategies...am I asking for the moon? |
Chiral is Tyler Durden |
|
|
 |
 FDAXHunter
|
Founding Member |
Total Posts: 8372 |
Joined: Mar 2004 |
|
|
As I said previously, this has been done pretty much 10 years ago. Back then most people decided that it doesn't really work in practice like it ought in theory, along with the whole ANN scene. Welcome to 1995. |
The Figs Protocol. |
|
 |
 NIP247
|
|
Total Posts: 559 |
Joined: Feb 2005 |
|
|
FDAX>> Does that apply for the whole concept of "universal" computational algorithms? (Universal portfolios, universalization of provably profitable trading rules etc.)
|
On your straddle, done on the puts, working the calls... |
|
|
 |
 kr
|
Founding MemberNP Raider |
Total Posts: 3561 |
Joined: Apr 2004 |
|
|
listen, I think it's an ambitious goal and appreciate the ideas, but I don't know how this is really supposed to work. I am just thinking of the first question about strategies - how do you classify, differentiate? Can it be done in a sufficiently systematic manner to split apart the winners from the losers? I was just looking at the venture capital fund returns this weekend, looking for familiar names, "quality" names... didn't actually run the statistics but it seemed like the correlation was between size of fund and returns (big = bad).
Really wish I knew where to do more reading on strategies - how to be really systematic about it. Everything I do in this space for single-name risks is very custom-tailored... which is not great b/c you don't have a good idea of the change in Sharpe vs. your alternatives. That's just a one-dimensional example. Anyhow, there's much more - and we've talked about it in this thread, which is good. But how about the literature - is there anything there which is good at all?
Where I know this stuff - in HY - the approach is usually: "Don't be fooled by this. Don't take that chance. Ratios shouldn't be in this region." i.e. exclude all the crap where you will get your head blown off. And yet this is not too different from the no-arbitrage idea - it's not easy to get paid for not taking risks, and the establishment of benchmarks just focuses the mind on what stats need to be manipulated. I wish I had something qualitative, stuff about rotation, changing your risk profile, etc. In my head I can think of market implied vols. vs. perceived expected vols, and how those connect with corp strategy, but I am looking for something more multidimensional... |
my bank got pwnd |
|
 |
 FDAXHunter
|
Founding Member |
Total Posts: 8372 |
Joined: Mar 2004 |
|
|
NIP247: Does that apply for the whole concept of "universal" computational algorithms? (Universal portfolios, universalization of provably profitable trading rules etc.)
IMHO it specifically applies to data mining all willy-nilly with GAs.
I'm not wild about GAs in general, as I've seen all attempts to lead them towards something useful ecclipsed by more specialized techniques. They are universal true enough, but exactly therein lies the problem: They are so universal that you can't really do anything specific with them.
Other "universal algorithms" would suffer from the same problem. |
The Figs Protocol. |
|
|
 |
 NIP247
|
|
Total Posts: 559 |
Joined: Feb 2005 |
|
| |
 |
 FDAXHunter
|
Founding Member |
Total Posts: 8372 |
Joined: Mar 2004 |
|
|
I haven't read that paper in it's entirety. However, I don't see how that is "universal", as it seems to deal specifically with (Markov) processes that have a fixed (albeit unknown) behavior? |
The Figs Protocol. |
|
|
 |
 Nonius
|
Founding MemberNonius Unbound |
Total Posts: 12808 |
Joined: Mar 2004 |
|
|
there are a number of large funds that mine for opportunities with algos (this goes way back and I'm sure Thorp, Simons, and a bnch of others were doing this a long long long time ago); many funds are quant in orientation, that is, from an opportunity-discovery perspective...not necessarily GAs, but algos nonetheless...I couldn't give a shit about GAs in particular...there are many ways of searching for patterns that are not simply iterating through all possibilities. the problem, as I see it, is that the dimensionality of iterating through everything is too high and covers securities, model parameters, mining choices, and what element of the distribution one is trying to predict...thus, jsut as one would not price complex derivative on 100 underlyings with anything other than something like MC, I would think that some sort of random+intelligent/intuitive mining algo would make sense. |
Chiral is Tyler Durden |
|
 |
 NIP247
|
|
Total Posts: 559 |
Joined: Feb 2005 |
|
|
In Cover's specific terms, a Universal Portfolio is a portfolio that is continuously rebalanced according to some rule to specifically beat the best instrument of the universe in hindsight (in absence of transaction costs, spreads, frictions, etc )
The paper about trading markovian price models refers to Covers' work as the worst-case scenario, i.e. when you really have no clue about the distribution. Kakade adds some restrictions and "derive" (from what I've understood) the optimal rebalancing rules that should hold iif the one of the models he uses are profitable. So he refers to covers work when he talks (in the abstract) of "universally profitable" strategy.
|
On your straddle, done on the puts, working the calls... |
|
|
 |
 FDAXHunter
|
Founding Member |
Total Posts: 8372 |
Joined: Mar 2004 |
|
|
Okay, I misunderstood the usage of the term universal then. But if that's your definition, I don't see what the question is? I can come up with an infinite number of strategies that are profitable in hindsight, the point is not to come up with many, but to come up with few that actually continue to work. We had a thread on this somewhere. |
The Figs Protocol. |
|
 |
 Nonius
|
Founding MemberNonius Unbound |
Total Posts: 12808 |
Joined: Mar 2004 |
|
|
Really wish I knew where to do more reading on strategies - how to be really systematic about it. Everything I do in this space for single-name risks is very custom-tailored... which is not great b/c you don't have a good idea of the change in Sharpe vs. your alternatives. That's just a one-dimensional example. Anyhow, there's much more - and we've talked about it in this thread, which is good. But how about the literature - is there anything there which is good at all?
I don't know. I would guess that the best funds that use it probably are not going to be keen on writing articles about it. the only fund manager that I know of who writes articles about what he thinks about is Thorp. anyway, perhaps you can take my word on it that...<<sorry had to delete that>>. they are all >5bill in capital, and they put on lots and lots and lots of trades.....anyway, the other funds that are pontificating on global economics and issuer-wise sheah are ok, but, at least from my perspective, there is a certain ad-hocness to it. One large European fund is the exception...I like those guys, and they have relatively few strategies spawned by trader's instinct coupled with thourough thought on what will happen in Japan, Europe, the US, etc.... |
Chiral is Tyler Durden |
|
|
 |
 kr
|
Founding MemberNP Raider |
Total Posts: 3561 |
Joined: Apr 2004 |
|
|
yeah - but is there absolutely no theory at all? I've been sort of thinking that a Michael Porter-type analysis could be applied, i.e. 'analysis of competitive strategies' or something like that... you see HFs starting to show up on the horizon, providing liquidity, usurping more traditional investors in the asset - almost like macro trading. There are lots of books on Portfolio Management, but I haven't found anything profound. I'd rather have 'Beauty Contest' type stuff, but from my vantage point I don't have a clear picture of all the new agents and what they hope to achieve - it's not at all my expertise (though certainly you've thought about this stuff).
re: ad-hoc, I know what you're thinking here, that's something I think about. This was one of the key things I asked my latest target: "How does your approach interface with the operation's macro view? Does it have to be consistent?" I was told at another firm (the one with the embattled CEO) that their macro views are not even internally consistent, i.e. their economists don't always agree, and it's ok. But if you're putting on a trade, there's no point in going long this morning and short this afternoon unless you want to just give money away or you are VERY MINDFUL about basis issues.
re: universal portfolio crap, even then you have to make assumptions about adjustment speed and stationarity which will always kill you. I think of plain old daytrading here, pyramiding up good positions and then scaling them back - if you really knew how to do this with strategies, and could operationally carry it off, you'd be a genius in my book. But I think that is almost entirely uncharted territory. |
my bank got pwnd |
|
 |
 Nonius
|
Founding MemberNonius Unbound |
Total Posts: 12808 |
Joined: Mar 2004 |
|
| |
 |
 apine
|
|
Total Posts: 1009 |
Joined: Jun 2004 |
|
|
salomon (i guess now smith barney) has a decision tree framework for some of their quant work for deciding on how to invest/allocate. i think it is pretty interesting from a far back perspective. i have not read the documents. but it might have potential to deal with the issues of perspective, stationarity, etc that kr and others have brought up (state variables for the comp sci crowd). the idea being that your agents/algos wax and wane as the state variables change. some of the models are dusted off when the yield curve gets steep and the fed is tightening and deficits are high. others when the curve is flat, credit spreads are wide and equities are in a down trend, etc.
anyone dealt with these thngs? |
Too many people make decisions based on outcomes rather than process. -- Paul DePodesta |
|
 |
 JabairuStork
|
Beat Box King |
Total Posts: 970 |
Joined: May 2004 |
|
|
Citi runs a little bit of prop money off of something that sounds like what apine describes. I couldn't help but think it sounded like a data mining exercise, e.g. 2-10 UST spread as a function of USD/MXF rate. But maybe there is something to it. |
|
|
|
 |
 Nonius
|
Founding MemberNonius Unbound |
Total Posts: 12808 |
Joined: Mar 2004 |
|
|
I think the term I see a lot in the lit/web is "data snooping". |
Chiral is Tyler Durden |
|
 |
| |
 |
 Nonius
|
Founding MemberNonius Unbound |
Total Posts: 12808 |
Joined: Mar 2004 |
|
|
markets are not efficient.
there are nuggets of gold.
its a numbers game.
a lot of the mining needs a bunch of machines.
I think. |
Chiral is Tyler Durden |
|
 |
 Bytes32
|
|
Total Posts: 8 |
Joined: Aug 2021 |
|
|
wow... you guys were talking about genetic algorithms, ANNs, SVMs, decision trees etc. in 2005! Really puts things in perspective. |
|
|
|
 |
 Maggette
|
|
Total Posts: 1350 |
Joined: Jun 2007 |
|
|
Yeah. Finance people, among some signal processing/control engineers were really early adopters of ML topics.
edit: at a closer look, Nonius also introduced soime kind of "human in the loop" ML approach here.
And damn I miss that kr guy. |
Ich kam hierher und sah dich und deine Leute lächeln,
und sagte mir: Maggette, scheiss auf den small talk,
lass lieber deine Fäuste sprechen...
|
|
 |