Daily Archives: December 22, 2011

Twitter Updates for 2011-12-22

  • Looking for a quant job? You may want to read this to filter out the bad ones: http://t.co/TTOeF1Qp #
  • Java vs C++ performance debate: http://t.co/1cGzcv2P #
  • New blog post: How to avoid the quant/developer black hole http://t.co/Gp5gxMz1 #
  • New blog post: IS the only disadvantage of ZeroMQ which does not give you low latency for HFT? http://t.co/OeQbJY41 #
  • New blog post: More info on Java vs C++ debate for HFT performance with Java JIT compiler including IBM Websphere http://t.co/CC752LVt #
  • http://t.co/LJuftoTP The question is, is 0MQ high performance compared to what others are talking about? From my perspective, I would go… #
  • New blog post: Bank of India -Excellent Short Sell Opportunity http://t.co/XwGCK9wH #
  • New blog post: Jointly modelling interest rates and equity returns with these quant analytics http://t.co/g3mpKk74 #
  • New blog post: Logistic regression for quant analytics http://t.co/wA9KkjWR #
  • New blog post: Quant development: HPCC Systems from LexisNexis Breaks World Record on Terasort Benchmark http://t.co/N1XtscIC #
  • http://t.co/2DlY1PQM Uh. I am not trying go one way or another but has anyone ever looked at 0mq?… #
  • New blog post: Choosing the Dell Server that's Right for You for HFT and quant development http://t.co/m7ozYWeq #
NOTE I now post my TRADING ALERTS into my personal FACEBOOK ACCOUNT and TWITTER. Don't worry as I don't post stupid cat videos or what I eat!

Why use other services for real time market data when you can get Yahoo Finance for $14/month? Or Google FInance?

Why use other services for real time market data when you can get Yahoo Finance for $14/month? Or Google FInance?

Fck! What a mess some of these market data providers are. They are either slow, prepay this, or want to know your life. My answer is … screw it. I am going with Yahoo Finance Real Time for really cheap! $14 per month. Get a connector going for my platform and I should be golden.

http://billing.finance.yahoo.com/realtime_quotes/signup

or maybe Google Finance

http://www.milliondollarjourney.com/google-offers-free-real-time-quotes.htm

NOTE I now post my TRADING ALERTS into my personal FACEBOOK ACCOUNT and TWITTER. Don't worry as I don't post stupid cat videos or what I eat!

Why is Esignal so expensive and with this higway robbery for API connections if it does not work? Does not make sense?

Why is Esignal so expensive and with this higway robbery for API connections if it does not work? Does not make sense?

Here is another chat with an Esignal rep:

 

 

 

Scott – Interactive Data has joined this session!

 

Connected with Scott – Interactive Data. Your Reference Number for this chat is 546667.

 

Scott – Interactive Data: Hello, thanks for using LiveRep.

 

bryan: hey scott

 

Scott – Interactive Data: HI

 

Scott – Interactive Data: Before we get started,may I ask how you found out about our service?

 

bryan: i am interested in the 30 day trial of esignal pro with a subscription ?

 

Scott – Interactive Data: Also, please feel free to contact me direct at 1-800-322-1819 OR 1-510-723-1629 anytime during this chat OR I can call you at any phone you specify.

 

Scott – Interactive Data: OK great

 

bryan: also I would need eSignal API entitlement

 

Scott – Interactive Data: I see

 

bryan: where do i go online and how long to activate ?

 

Scott – Interactive Data: Have you ever had our service before?

 

bryan: o

 

bryan: no

 

Scott – Interactive Data: Are you interested in stocks or futures?

 

Scott – Interactive Data: Did you want me to call you?

 

bryan: and forex but stocks for now ?

 

bryan: no need to call i will fill out online

 

Scott – Interactive Data: Just stocks?

 

bryan: for now but would like to try forex with futures if possible?

 

Scott – Interactive Data: Cost of eSignal is $154.00 a month,, plus a $18.00 monthly U.S. stock exchange fee. Total is $172.00 a month. Plus there is a one time $195.00 ActiveX API fee.

 

bryan: what about a trial for both? i want to ensure works with my platform?

 

Scott – Interactive Data: What platform are you using? it may already work with our data.

 

bryan: xxx here are the instructions work with esignal

 

 

Scott – Interactive Data: Are you using Ensign , or Metastock or OmniTrader?

 

bryan: read the link XXX

 

Scott – Interactive Data: So you’re developing your own application?

 

bryan: good gawd, read the link

 

bryan: it is open source with 18000 downloads

 

Scott – Interactive Data: OK

 

bryan: as said, i would like a trial to see it if it connects

 

Scott – Interactive Data: Sorry, we only offer a 30 day trial money back guarantee, if you’re not happy, we’ll give you a full refund less the $18.00 exchange fee and the $195.00 ActiveX API fee.

 

bryan: so i have to pay upfront ?

 

Scott – Interactive Data: AGain , $18.00 and $195.00 is not refundable.

 

Scott – Interactive Data: Yes

 

bryan: i don’t think so

 

bryan: thansk anyhow

 

Scott – Interactive Data: Good luck

 

Print Copy Exit
 

Scott – Interactive Data has joined this session!

 

Connected with Scott – Interactive Data. Your Reference Number for this chat is 546667.

 

Scott – Interactive Data: Hello, thanks for using LiveRep.

 

bryan: hey scott

 

Scott – Interactive Data: HI

 

Scott – Interactive Data: Before we get started,may I ask how you found out about our service?

 

bryan: i am interested in the 30 day trial of esignal pro with a subscription ?

 

Scott – Interactive Data: Also, please feel free to contact me direct at 1-800-322-1819 OR 1-510-723-1629 anytime during this chat OR I can call you at any phone you specify.

 

Scott – Interactive Data: OK great

 

bryan: also I would need eSignal API entitlement

 

Scott – Interactive Data: I see

 

bryan: where do i go online and how long to activate ?

 

Scott – Interactive Data: Have you ever had our service before?

 

bryan: o

 

bryan: no

 

Scott – Interactive Data: Are you interested in stocks or futures?

 

Scott – Interactive Data: Did you want me to call you?

 

bryan: and forex but stocks for now ?

 

bryan: no need to call i will fill out online

 

Scott – Interactive Data: Just stocks?

 

bryan: for now but would like to try forex with futures if possible?

 

Scott – Interactive Data: Cost of eSignal is $154.00 a month,, plus a $18.00 monthly U.S. stock exchange fee. Total is $172.00 a month. Plus there is a one time $195.00 ActiveX API fee.

 

bryan: what about a trial for both? i want to ensure works with my platform?

 

Scott – Interactive Data: What platform are you using? it may already work with our data.

 

bryan: xxx here are the instructions work with esignal

 

 

Scott – Interactive Data: Are you using Ensign , or Metastock or OmniTrader?

 

bryan: read the link XXX

 

Scott – Interactive Data: So you’re developing your own application?

 

bryan: good gawd, read the link

 

bryan: it is open source with 18000 downloads

 

Scott – Interactive Data: OK

 

bryan: as said, i would like a trial to see it if it connects

 

Scott – Interactive Data: Sorry, we only offer a 30 day trial money back guarantee, if you’re not happy, we’ll give you a full refund less the $18.00 exchange fee and the $195.00 ActiveX API fee.

 

bryan: so i have to pay upfront ?

 

Scott – Interactive Data: AGain , $18.00 and $195.00 is not refundable.

 

Scott – Interactive Data: Yes

 

bryan: i don’t think so

 

bryan: thansk anyhow

 

Scott – Interactive Data: Good luck

 

 

NOTE I now post my TRADING ALERTS into my personal FACEBOOK ACCOUNT and TWITTER. Don't worry as I don't post stupid cat videos or what I eat!

How to avoid the quant/developer black hole

How to avoid the quant/developer black hole

This was sent to me by someone on my Google +1

http://quantjob.blogspot.com/2011/12/how-to-avoid-quantdeveloper-black-hole.html

This article is actually quite good to filter out the shi*tty jobs out there. Interesting read.

However I do not agree on the PERL argument.

NOTE I now post my TRADING ALERTS into my personal FACEBOOK ACCOUNT and TWITTER. Don't worry as I don't post stupid cat videos or what I eat!

IS the only disadvantage of ZeroMQ which does not give you low latency for HFT?

IS the only disadvantage of ZeroMQ which does not give you low latency for HFT?

 

Being part of Linked In group discussion, this came up:

0mq has one disadvantage IMO (I can be wrong because I am not an expert in 0mq). It can not utilize “user space TCP/IP” stack implementations of a a modern 10G or IB interfaces which will get you latency in low microseconds.

True or false?

NOTE I now post my TRADING ALERTS into my personal FACEBOOK ACCOUNT and TWITTER. Don't worry as I don't post stupid cat videos or what I eat!

More info on Java vs C++ debate for HFT performance with Java JIT compiler including IBM Websphere

More info on Java vs C++ debate for HFT performance with Java JIT compiler including IBM Websphere

It looks like it is part of IBM Websphere solution but here are some of other links:

http://publib.boulder.ibm.com/infocenter/realtime/v1r0/index.jsp?topic=%2Fcom.ibm.rt.doc.10%2Frealtime%2Frt_jit.html

http://www.gnu.org/software/dotgnu/libjit-doc/libjit_3.html

http://docs.oracle.com/javase/1.3/docs/tooldocs/win32/java.html

http://docs.oracle.com/javase/6/docs/api/java/lang/Compiler.html

 

NOTE I now post my TRADING ALERTS into my personal FACEBOOK ACCOUNT and TWITTER. Don't worry as I don't post stupid cat videos or what I eat!

Quant development: Links on JNI for this wierd debate on who is faster C++ or Java. Needed to know for HFT uses

Quant development: Links on JNI for this wierd debate on who is faster C++ or Java. Needed to know for HFT uses

I keep reading about this C++ versus Java performance debate. They always go one about JNI so here are some better articles about it the use of it.

http://www.javaworld.com/javaworld/javatips/jw-javatip17.html

http://home.pacifier.com/~mmead/cs510jip/jni/

Also, other mention about a better JVM called Rockit. Don’t look at me on this debate as I already use C++ and C#. I got kind of tired of where Java is going these days.

 

NOTE I now post my TRADING ALERTS into my personal FACEBOOK ACCOUNT and TWITTER. Don't worry as I don't post stupid cat videos or what I eat!

Bank of India -Excellent Short Sell Opportunity


Bank of India -Excellent Short Sell Opportunity

Bank of India is giving a good short term short sell opportunity. The stock has broken out downwards from a rising wedge pattern on its weekly charts, as on 16-12-11, with a rise in volume. Trad…

 

ya buddy, u r right…

 

 

NOTE I now post my TRADING ALERTS into my personal FACEBOOK ACCOUNT and TWITTER. Don't worry as I don't post stupid cat videos or what I eat!

Jointly modelling interest rates and equity returns with these quant analytics

Jointly modelling interest rates and equity returns with these quant analytics

Does anybody know of a parsimonious model to jointly generate timeseries of asset classes (say monthly returns of equity indices and government bonds), that allows for varying correlation between the two asset classes over time? I am looking for something that can be calibrated against a longer period (say 70 years) spanning times of falling rates as well as times of rising rates, capturing both environments.

 

 

==

First you need invariants. For equities that’s log returns, for government bonds its changes in YTM. On the other hand, if you have a return series for bonds in addition to yields, you could probably also throw that in with log changes. Since you’re looking at a long period of time, you probably should also consider throwing in something like log changes in inflation and (since you’re looking at monthly data) you may want to throw in something from the real side of the economy like log changes in IP.

For the sake of simplicity, let’s you only are looking at these four/five things. Now you could use the changes I mentioned above, but it turns out that what’s really important is the cointegrating relationship between inflation and interest rates (and lagged values of interest rates). So you should probably estimate a multivariate ornstein-uhlenbeck process on the levels of the above series. That handles the autocorrelation and multivariate cointegration. However, it doesn’t handle contemporaneous correlations and time-varying volatilities, which is where you would be concerned most about time variation. To account for time-varying volatilities, fit a Garch model to the residuals. Then I would scale the residuals by the conditional standard deviation. Next, convert the scaled residuals to a copula and fit a DCC copula (there’s a matlab package for dynamic copulas, but I’m not wildly enthusiastic about the way it is structured). DCC Copulas allow for time-varying correlations that you’re concerned about.

You could also include things like Markov regime-switching (there’s a Matlab package for it that I am enthusiastic about).

One recent concern is modeling the zero lower bound on rates. This paper provides steps in the right direction:
http://cama.anu.edu.au/Working%20Papers/Papers/2011/Krippner362011.pdf

 

==

It is interesting but if you first estimate OR process than estimate multivariate GARCH and DCC copula I wonder how stable is the whole setting. Is the GARCH independent on OR?

Apart from that, I would like to follow no with a question on how should the bond invariants be treated. Say I would like to model equity together with government and corporate bonds. My understanding is that I should start with swap curve where I need to reduce the dimension to 1-3 factors using PCA on covariance matrix. Then I would probably proceed with PCA on each of credit spread curves to get some invariariants there as well. Now when I have invariants for swap curve, individual credit curves and equity (log returns) I would probably add inflation, IP or sentiment and maybe something like VIX and jointly estimate OR process. Do i have it right or is there some other way to deal with different types of bonds?

 

==

For such long time span: 70 years it is very difficult to define invariants.
If you have sufficiently long time series I would try to be “nonparametric”, instead of fitting complicated models. I think that the entropy pooling articles give good suggestions on how to generate scenarios.

Moreover, you should consider different “states” of the economy, like: business cycle, monetary cycle, …

 

==

what abouti)semi-parametric approach, which fit empirical distribution for two variables of equity index and bond; then fit a bivariate copula. I think the challenge is to choose the right copula; ii) Alternatively, you could think of “mixture” approach, different from the traditional mixture distribution, use GARCH to fit the central part of joint distribution and use copula to fit the tails. And in this way, you will need to identify a “distance” measure to split central(GARCH) and tail(copula).
E.g. http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1260228

 

 

NOTE I now post my TRADING ALERTS into my personal FACEBOOK ACCOUNT and TWITTER. Don't worry as I don't post stupid cat videos or what I eat!

Logistic regression for quant analytics

Logistic regression for quant analytics

While selecting the model for logistic regression, it is observed that the removal of particular variable x from the model improves the accuracy significantly but does not reduce the residual deviance as compared to the case in which all variables are considered. How to tackle this?

 

 

 

Did you do the Likelihood ratio test for both model? G(diff)= G(initial)-G(reduced) as well as Dof, and using Chi2 to test if they are significant different. Also proceed with caution that X variables may have multicollinearity.

Hope this helps

 

thnaks for your reply. i think there is multicollinearity problem. i did the G(diff) and it is high.

 

NOTE I now post my TRADING ALERTS into my personal FACEBOOK ACCOUNT and TWITTER. Don't worry as I don't post stupid cat videos or what I eat!