Serialization delays

There is a lot of innovation and investment in Technology in Finance nowadays. There are submillisecond advancements being made to gain an upper hand in this world. Seems like the time when Google posting the time of the search on the website to pull in people has ended. Now speed has found a new home.

House Republicans save the Economy, Obama fails to show leadership

BarackObama blog "save the economy"

It is disappointing that while House Republicans were able to listen to the people in their constituencies, listen to ex Federal Reserve Board members how the bailout plan of giving money to banks who have paid out more than 3 B to just their CEOs in the last 3 years while losign billions for shareholders and making billions off the people who have taken loans from them, while House Republicans refused to bullied by their leaders by the traders on the stock market, Sen. Obama did nto show ANY of the leadership and change he has so promised. We urge you Sen./ Obama to take some tiem and talk to people who have supporeted you whether they feel money should be GIVEN ( in whatever form ) to the institutions and the traders who have made hundreds of billions while creating the problem. People like Bill GRoss had pointed out significant possibility of this happening while the need to make mroe profits and leverage the money more drove companies to ignore all this and forge ahead.

Thankfully to House Republicans Sen. Obama you still have time, pleasetry to work out a much better solution o this than what understandably biased Treasury Secretary has proposed so far. Please look at Warren Buffett's way of capital injectio. If Goldman Sachs ( Paulson's old company ) had to borrow from Buffett with such an enticign deal the U.S. taxpayer should not be forced to give money getting back this almost worthless bad debt! Please at least get us a deal like Buffett.

Algorithmic trading

Recently I have seen a lot of people from IITs enter the finance industry in Algorithmic trading. I don't think the industry is saturated at all, but statistics and computer science methods are not as applicable to this form of trading as can be made to believe. The fact is that more than 60 percent of the trading on most products is done by at most 4 to 5 traders who can control the combined performance of all the other market participants and easily squeeze them by sort of collusion. and collusion happens almost inadvertently if you model your adversary and try to estimate combined positions changes for other members. Basically I wanted to say that the insample and out of sample performance of all computer science methods in high frequency finance is going to be abysmal due to some of the facts above ad just the number of people trying to take so much risk for so little profit, that only the adverse fills are exaggerated.

hillclimb based regression

Starting with Y and X1 to Xn, find the Xi which is most correlated and eliminate it's effect from Y and other Xs left. Then repeat the process and at the final step compute the coefficient of the X as per the one variable regression problem. Then returning back the recursion keep adding coeffienmnts for each variable using the one variable regression formula and add the parts which where removed from the variables. A better discussion of it can be found in a report by Taranbir Singh and Rakesh Garwal.

empirical stuff about fast moving products

A simple statistic of a fast moving product is the AR coefficint in the returns series, when snapped every 2 seconds, also a quantile quantile transition amtrix would have a high values along the ends of the principal diagonal. Rakesh Kumar attested the fact about nasdaq 100 futures, wich ost will accept as having a much lower beta than most stocks. interestingly about nasdaq 100 i was surprised to see that the holdings of apple are much more than that of dell and intel. must say they have doen well for themselves. i fell qqqq has fallen much more than it should and when the US is growing at 6% nasdaq being at the same level as it was 16 months ago does not seem justfied.

linear regression continued...

Although lacking many of the statistical properties of regression.. the aforementioned method of computing the linear regression coefficients can be very easily optimized to minimize the function Sum ( i to N ) ( ( yi - xi ^T beta ) * f ( y ) ). where f( y ) can be a sort of funtion specifying how much importance to give to a point in the sample. Many applicatons of regression may require f(y) to be say, a step function of the absolute value of y or a positively biased sampler or a sampling proportional to the absolute value or maybe a square of it.
Mansi pointed it out to me that linear regression if doen with all the variables gives the same coefficients as a hillclimb verion of it ( in the newest post )

linear regression versus a system of equations

It turns out that the closed form linear regression solution of Y predicted with X with coeffs beta = (X^tX)^-1X^tY is the same as the one that would be obtained by differentiating the sum of squared deviations with each parameter in beta and solving the system of linear equations... courtesy Rakesh Kumar

bond maths contd....

the term structure found can also be equa to the zero - coupon bond yield.
choos ean nterest rate model that is consistent with the yield curve obtained from market prices of bonds. To calclate the zero coupon yield curve implied by an interest rate.

Pricing interest Rate Sensitive Securities I

In case of pricing stocks we use a tree assuming the markovian model. This does not work n case of fixed income securities like bonds, since th amount to be paid at the end is fixed. What does determine the present value of a fixed income secuirty is an ineterst rate model. The interest rate model can be assumed to be markovian.
For veery interest rate model there is an equivalent term structure model that can be observed only in the market. The probability of rise and fall from a node in the interest rate tree is assumed to be (1) independent of time and state (2) dependent on state (3) dependent on time (4) dependent on time and state. Ingersoll and Ross methods capture the term structure the best but are computationaly difficult.
The main idea is that given the sequence of future discountings, the expected present value of asecurity can be calculated. In equity market the lognormal model is standard. In FIRC market there is no standard interest rate tree model. Each business house uses its own peoprietary MODEL.

Trend catching

I was reading this article in statistics and empirical finance, which talks of asymmetry in time horizons of investors. These things motivate many game theoretic strategies of automated trading. If we could generate a description of how different investor classes react to information on wall street then we can think of trading strategies which, looking at this and the current state of order books, can find a strategy of optimally placing requests to catch the majority of any trend.
It is clear that trends in price start after some external impulse directly related to the company. Hence understanding what information is relevant (broadly) is not difficult. There has to be a notion of automated/human corroborated interpretation of information. Often technical information, like moving averages etc are very easily discovered. Based on the different horizons and different investor class contribution to overall market movement, one can devise games to win in this model.