Forums  > Trading  > quant (algorithmic) macro models  
     
Page 1 of 1
Display using:  

TheSheriff


Total Posts: 3
Joined: Apr 2009
 
Posted: 2009-04-28 04:17
Hello all,

I have a back-burner type project/issue that I've been pondering for a while, and would like to pose it to the forum. By way of background, I'm a (former) swaptions trader (4 years), and now I'm working on some entrepreneurial ventures in the quant equity space.

I'm generally familiar with models for algorithmic long/short equity trading over medium to long time horizons (nothing high frequency, think 1m to 1y holding periods.) I'm also, through my former job, familiar with what I would call quantitative (but not fully algorithmic) methods of trading US fixed income instruments. These would include trading vol butterflies by watching correlations/regressions vs whats going on with the interest rate curve, trading butterflies on the curve, trading spread models vs the absolute level of rates and implied vol.

All of these quantitative methods in fixed income, while quantitative, were definitely not algorithmic. This is understandable in the execution sense, since its an OTC market, but they also were not algorithmic in the sense of being automated on the analysis side, because they always tended to include a priori views contingent on external factors.

Now, what I've been mulling over is whether there is a general consensus on how to approach the problem of automating macro trading across asset classes. For instance, factor models make sense for equities but not so much for bonds since it only takes a few parameters to completely determine the bonds present value (if you are talking about treasuries, treasury futures, swaps, etc. Corporate bonds could be a different issue.)

But when incorporating other asset classes such as FX, commodities, etc are where I have trouble getting a grasp on what if anything is considered standard or reasonable approaches. In some asset classes, like commodities, some of the problem seems to be the almost pure supply and demand nature of the instrument, and perhaps the lack of consensus on how to put even begin to theoretically value the assets in simplified or idealized settings.

A few ideas I've had are, in FX for instance, treating each tradeable currency like an individual stock. To me this approach has the advantage that you could port many of the same techniques from factor based quant equity strategies (including now having a large number of comparable inputs/statistics), and it even has an analogous situation of having a "theoretical" (to at least some extent) accepted method of valuing the instrument (PPP equivalence), and then using whatever analysis, fundamental factors, momentum factors, etc, you want to generate a long/short portfolio. One problem, among others, is the tradeable universe if very small.

Another idea I've had is to simply gather as much time series data as possible across FX, commodities, rates, equities, and economic data, and try to do a pairwise search for co-integrated variables and then sit down with each result and determine what, if any, if the accepted governing macroeconomic relation. This seems extremely tedious and time intensive, and a bit like re-inventing the wheel. I fear it also points out the problem that true macro trading needs to be built on some pre-conceived notion of what the various relationships between macro variables are. I hope that, like the case for equities, its possible to ferret out a tradeable relationship using, for example, a model with linear combination of some not too large number of factors, without having that model reflect in a pre-constructed way a consensus method of valuing whatever you are looking at.

Does anyone out there work in or look at this area? Has anyone put any thought or effort into building algorithmic cross-asset class macro models? Can anyone suggest a book or paper that discusses whats reasonable or unreasonable?

Path Integral


Total Posts: 441
Joined: Dec 2004
 
Posted: 2009-04-28 14:27
For FX you might have a look at work from Paul Daoust on intrinsic currency valuation.

Too much hanky panky makes Fannie bleed

jungle
Chief Rhythm Officer
CSD LLC
Total Posts: 3169
Joined: Jul 2004
 
Posted: 2009-04-28 14:48
Oodles of stuff out there on trading [whatever] based on economic data, you just need to look for it.

"according to the banks, some assets "no longer correspond to their present commercial strategy and... should therefore be externalised." BAFIN

briant57


Total Posts: 98
Joined: Feb 2009
 
Posted: 2009-04-28 15:36
TheSheriff,

I picked up a copy of Intermarket Trading Strategies by Katsanos, a new book out Feb 2009. There are some good chapters on how to select inputs to your model, and at the least it would help you generate more ideas.

Amazon- http://tinyurl.com/cjn8tx

I have recently started putting together a model to trade the /ES based on movements in a few selected commodity, interest rate, and FX futures contracts. I have not incorporated economic data into my model, I am focused on a shorter term that what you are looking for (+/- one day holding period). Once I am finished, I hope to trade the independent variables in a similar fashion essentially creating a basket.
If you are interested in talking further drop me an email.


FDAXHunter
Founding Member

Total Posts: 8367
Joined: Mar 2004
 
Posted: 2009-04-28 20:48
And you want to replace "algorithmic" with "systematic", otherwise everyone will laugh at you.

The Figs Protocol.

TheSheriff


Total Posts: 3
Joined: Apr 2009
 
Posted: 2009-04-28 22:43

Really? I view "algorithmic" as more constrained than "systematic". I could say I had a system when I traded swaptions, but steps in the system might be something as generic as "determine when market consensus is heavily weighted toward some trade and take the other side". to me algorithmic implies you can implement it in code, whereas systematic could leave some of the steps in the code to processes inside the human mind. plus, i've used the term algorithmic with some pretty experienced investors and i didn't get laughed out of the office. (yet!)

Anyone else have an opinion?

 

 


jungle
Chief Rhythm Officer
CSD LLC
Total Posts: 3169
Joined: Jul 2004
 
Posted: 2009-04-28 22:59
"Algorithmic trading" means using algorithms to improve execution, which is certainly not the same thing as systematic trading. 

"according to the banks, some assets "no longer correspond to their present commercial strategy and... should therefore be externalised." BAFIN

FDAXHunter
Founding Member

Total Posts: 8367
Joined: Mar 2004
 
Posted: 2009-04-28 23:02
Well, I laughed when I read the thread title.

Algorithmic Trading relates specifically to the automated processing of orders to minimize market impact. It is not systematic trading. Of course, there's a lot people who don't know the difference between the two (Clueless people in finance?! Surely not!!). I'm just saying: in the right circles, that'll get you laughed at.

Edit: Crossed posts with jungle.

The Figs Protocol.

TheSheriff


Total Posts: 3
Joined: Apr 2009
 
Posted: 2009-04-28 23:10

Ah, I see. Yes, I know people who work to develop block trade execution algorithms and the like, so I could certainly see that.  Still, my goal being to develop trading "systems" that are completely automated from data acquisition to trade execution I think the term applies. Nonetheless, has anyone else out there anything to say about approaches to systematic macro strategies?

 

 


sfca


Total Posts: 904
Joined: May 2004
 
Posted: 2009-04-28 23:14
Sure, as long as you want to appear fragile.  That PPP thingy.  No go.  PPP is about transactions with goods in theory and is only now useful for classroom Econ 1 analysis.  Capitial markets completely overwhelm that impact.  While PPP may influence things decade to decade, your trades will not last that long.  It would be kind of like you taking continental drift into consideration when planning your next car trip. 

FDAXHunter
Founding Member

Total Posts: 8367
Joined: Mar 2004
 
Posted: 2009-04-28 23:17
Re: Algorithmic Trading. Well... think whatever you like, you're still wrong. Smiley

I won't be able to add anything to the systematic side of things with regard to macro trading. The consensus seems to be is that the context in which to interpret macroeconomic data evolves too fast (relative to the macroeconomic shifts) to be able to analyze it consistently using the same methods. (But we know the consensus tends to be wrong, right?).



The Figs Protocol.

sfca


Total Posts: 904
Joined: May 2004
 
Posted: 2009-04-28 23:25

The guy with the badge (Sheriff) might want to examine Wikipedia on the Lucas Critique. 

[edited for clarity]

 


jungle
Chief Rhythm Officer
CSD LLC
Total Posts: 3169
Joined: Jul 2004
 
Posted: 2009-04-28 23:58

Another idea I've had is to simply gather as much time series data as possible across FX, commodities, rates, equities, and economic data, and try to do a pairwise search for co-integrated variables and then sit down with each result and determine what, if any, if the accepted governing macroeconomic relation. This seems extremely tedious and time intensive, and a bit like re-inventing the wheel. I fear it also points out the problem that true macro trading needs to be built on some pre-conceived notion of what the various relationships between macro variables are. I hope that, like the case for equities, its possible to ferret out a tradeable relationship using, for example, a model with linear combination of some not too large number of factors, without having that model reflect in a pre-constructed way a consensus method of valuing whatever you are looking at.

Isn't that just the old structural vs. empirical argument?  Macro models vs. VAR?  etc. 

I'm not inclined to say much more b/c it really seems from your post(s) that you haven't done much homework on any/all of this.


"...according to the banks, some assets 'no longer correspond to their present commercial strategy and... should therefore be externalised.'" BAFIN

sfca


Total Posts: 904
Joined: May 2004
 
Posted: 2009-04-29 00:21
Instead of the datamining exercise, perhaps he could start small.  I'd suggest that he look at the cds/corporate bond basis and how that has been impacted by macro factors, consumer confidence, international capital flows, and such and see if he can generate a predictive model of the basis.  He could incorporate the Merton model and whatever else might be entertaining.  

aaron


Total Posts: 746
Joined: Mar 2006
 
Posted: 2009-04-29 01:04

You mention the difference in quant equity versus systematic strategies in other markets, and attribute it to automated execution in equities. That's part of it, but not the entire story. For one thing, of course, you can automate execution in other markets.

But the big point is there is far more idiosyncratic risk in equities than any other market. A lot of quantitative strategies exploit that by putting on hundreds or thousands of positions with tiny positive Sharpe ratios. And the idiosyncratic risk is easily tradeable, there's lots of liquidity and it's easy to isolate.

In any other market, you're going to be faced with a relatively small number of factors to bet on. You're going to need higher Sharpe ratio strategies, because you won't be able to implement as many of them either at one time (fewer factors) or over time (less high-frequency liquidity), with as much independence.

The pure computer trading works best when averaging results over a large number of trades. When you get fewer trades, it makes more sense to investigate each one more carefully.

I'm not attacking your project, just pointing out that you may be stretching an analogy between quant equity and other markets.


MrMagoo


Total Posts: 192
Joined: Jan 2008
 
Posted: 2009-04-29 01:26
Two paths to trade with a systematic macro perspective :

1 - Hard and Very Time Consuming : Learn macroeconomic theory - the core stuff, forget about current academic research that focus on imperfections. Read mostly non-nonsense austrian theory papers, available at mises.org . The following paper may also helpful What do we know about Macroeconomics that Fisher and Wicksell did not?

keep in mind that (usually) good macroeconomic thinking is like wine, requires a maturity that comes after some years practicing.

To focus on systematic trading, my experience shows me that the key is to develop models for the asset risk premiums. Simply stated, when the economy goes under stress or recessions, the risk premiums rise sharply (that means Dividend/price ratio, Bond risk premiums, etc. ,all soar.).

Many models capture that (think of multifactor models), but it is a fact that risk premiums are time-varying, so any unconditional model (All APT-like, Barra stuff) is useless for short-term trading . Conditional models work, but belive me, they are very hard to understand, estimate and trade. To get very realistic you can try to model market expectations, but be beware not to lose your sanity. You should know *very well* at least macroeconomics and non-trivial econometrics, not to be trapped into false or nonsense conclusions. Also, linear models short-term conclusions are dangerous in a non-linear world. Rational expectations assumptions are also very dangerous in our bubbly age.

That gives you a small idea of the deep shit you should be stepping in, when trying to combine quant models with a more intuitive macro-systematic view.

2 - ALMOST Pure Quant Models : You can bypass most of the deeper macro understanding using vector auto-regressive macro forecasts (given you understood lucas critique and the notion of "market states" - think of hidden markov models here -), and try to tie that to multi-class asset prices. It works, but model building and trading this way will take you some time to fine tune. It also doesnt hurt to understand a bit of monetary and fiscal policy regimes. Think of politics as exogenous shocks.

There are thousands of ideas you can later add to a basic macro model, assuming you can build one sucessfully .

"One who says it can't be done should not interrupt the person doing it."

g.kapur


Total Posts: 14
Joined: Feb 2008
 
Posted: 2009-04-29 19:23
"Sure, as long as you want to appear fragile. That PPP thingy. No go. PPP is about transactions with goods in theory and is only now useful for classroom Econ 1 analysis. Capitial markets completely overwhelm that impact. While PPP may influence things decade to decade, your trades will not last that long. It would be kind of like you taking continental drift into consideration when planning your next car trip."

Can you explain this comment? PPP has performed pretty well in the past year, I believe. While the IR on trading a pure PPP strategy isn't amazing, it's not something you should totally scoff at: http://www.firstquadrant.com/downloads/A_Simple_Measure.pdf

Any strategy that makes money, makes money... I'm not sure I understand the distinction of decades versus other time periods + some of the economics 101 strategies are actually fairly strong soft hedges against carry trades.

One thing you should note, is that trading macroeconomic factors is different in developed and emerging markets (obviously). Certain factors, which are fairly intuitive, seem to work only in emerging markets (there's some literature on CA as an indicator in emerging markets). One could argue that there's a monetary policy shift of some sort, which might be part of what's going on... but I think it has to do with instability of emerging economies and other considerations.

GK

Mystery


Total Posts: 11
Joined: Aug 2007
 
Posted: 2009-04-29 21:01

True macro trading needs to be built on some pre-conceived notion of what the various relationships between macro variables are.

Very true.

And if you research further you will see that most asset classes correlate with a limited number of macro variables.

I would replace algorithmic by quantitative.


sfca


Total Posts: 904
Joined: May 2004
 
Posted: 2009-04-29 21:03
The trade weighted value of the dollar has appreciated about 14 percent from April 2008, after depreciating 9 percent from April 2007.  Relative price changes can not explain that.  These movements reflect expectations, central bank interventions and capital flows.  I think its best to focus on what is driving most of the market moves rather than focus on persistent but weaker influences.  Having said that, I agree completely with you that you should do any strategy that will make money for you.  If you have a way to make money using PPP, great!!!    

jungle
Chief Rhythm Officer
CSD LLC
Total Posts: 3169
Joined: Jul 2004
 
Posted: 2009-04-29 21:23

Problems with PPP strategies are the same as any other valuation-based strategy, i.e. length of half-life, possibility of "mis-valuation" persisting for a long time; possibility of getting stopped out before you realise the PNL; variance of PNL etc. 


"...according to the banks, some assets 'no longer correspond to their present commercial strategy and... should therefore be externalised.'" BAFIN

macrotrader


Total Posts: 353
Joined: May 2009
 
Posted: 2009-05-12 14:05
>> Another idea I've had is to simply gather as much time series data as possible across FX, commodities, rates, equities, and economic data, and try to do a pairwise search for co-integrated variables and then sit down with each result and determine what, if any, if the accepted governing macroeconomic relation.

I can tell you what the result is: if you add butter production in Bangladesh, U.S. cheese production, and sheep population in Bangladesh and the U.S. together you can explain 99% of the annual movements of the S&P 500 between 1983 and 1993.

http://nerdsonwallstreet.typepad.com/my_weblog/files/dataminejune_2000.pdf

Two companies who practice quantitative global macro are AQR and QFS. People from both companies have published papers.

A)
http://www.aqrcapital.com/Research/ValMom%20AMP_20080710.pdf
B)
Search SSRN + Sanford J. Grossman

Maltese


Total Posts: 51
Joined: Aug 2005
 
Posted: 2009-07-04 22:09
You might want to look at the Chicago Fed activity index as an example of factor macro model. It is probably would not be considered state of the art at this point.

Thought is only a flash between two long nights, but this flash is everything. Henri Poincare.

quantie


Total Posts: 885
Joined: Jun 2004
 
Posted: 2009-08-01 16:54
to add to jungle, PPP=pretty poor predictor. It works in the long run but in the long-run we are ...

stanley.oliver07


Total Posts: 2
Joined: Mar 2010
 
Posted: 2010-03-14 21:13
Hi - I'm generally interested in the same area of research although not into algo's. Just quit my day job at a hedge fund and have some free time incase you want to calliberate on some project from start to finish.

Jurassic


Total Posts: 161
Joined: Mar 2018
 
Posted: 2018-07-03 19:51
@MrMagoo

Whats a conditional factor model? Is there a paper on this?
Previous Thread :: Next Thread 
Page 1 of 1