Week 9 Update

Finished off the dark pool simulations and granger causality testing results. However needed to redo the granger causality results in order to account for endogeneity problems. First thought was to do generalized method of moments however Sanjay prefered machine learning techniques as these were less archaic.

Week 6 and 7 Semester 2

conducted dark pool simulations to find the optimal algorithm for dark pool allocations. This was done on Matlab

Week 5 Semester 2

This week I met with Sanjay and he suggested adding a norm squared term to the Liquidity Algorithmic-Trading Frequency regression model to reduce the overfitting in the model.

Also, he asked me to start conducting the dark pool simulation. In the dark pool simulation I will be simulating dark pools using a Poisson Distribution and I will have a Person class that will conduct a multi armed bandit algorithm in order to optimize allocation across dark pools in the minimal time.

Week 4, Semester 2

This week I finished up the Granger Causality Tests on the 20 stocks from the ASX. Unfortunately we could not conclude that the proportion of algorithmic traders granger causes liquidity changes.

Also, I finished writing up the chapter on Cointegration, Granger Causality and Spurious Regression.

Now I will just show the results to Sanjay and then move on to conducting dark pool simulation experiments.

Week 2, Semester 2

Currently I am writing the chapter on cointegration, granger causality test and spurious regression.

Also, I am doing experiments on stocks to test for causation relationship between algorithmic trading frequency and liquidity.

Week Starting 18.07.2011

Achievements for this week:

Experimented to verify the hypothesis that Algorithmic Trading affects liquidity positively. That is when Algorithmic Trading Frequency rises Liquidity rises and when Algorithmic Trading Frequency  falls so does Liquidity.

• Extracted records regarding the trades of 20 stocks for the three year duration of the data available.
• Used MATLAB to measure the algorithmic trading frequency of these stocks on randomly chosen days for 5 minute blocks. Plotted the results on a line graph.
• Used MATLAB to measure the bid ask spread of these stocks on the same days. Plotted the results on a line graph. The bid ask spread is an indicator of liquidity.
•  It was observed that there is a correlation between algorithmic trading falls so does liquidity and this seems to indicate that there is a possibility that algorithmic trading may cause it. However causation does not equal correlation and so the Granger Causality Test was used on R to verify causality. Preliminary tests indicate causality does exist. However further verification is required.

Week Starting Monday, 11 July 2011

Focus for this week: To replicate the experimental analysis from the paper 'Does Algorithmic Trading Improve Liquidity?' By Hendershott et al.

Achievements for Monday:
1. Made  the plan for how to replicate the experiment from the above paper.
The most important steps are as follows:
(i) Find the number of stocks and group them into quintiles by market capitalization.
(ii)  By quintiles, graph the number of algorithmic traders involved across the time period for which data is available.
(iii) By quintiles, graph quoted half spreads, quoted depth, and effective spread.
(iv) Perform the regression analysis of algorithmic trading on liquidity.
(v) Identify whether there is an effect on frequency of AT due to liquidity.
(vi) Find Instrument Variable.
2. Towards achieving the plan above, I have done the following on Monday: Wrote up and started executing the python scripts to list and count the number of stocks, the dates for which the data exists, and the mean market capitalization for each stock. These all allow me to progress towards achieving step (i). I also sent an email to my finance supervisor Andrew Lepone regarding the question of what instrument variable we may use in the context of ASX. Since the paper focuses on the NYSE and over there they used the introduction of the autoquote as the instrument variable-and we cant do that here.


Achievements for Tuesday
Today I wrote up and executed the python scripts to extract the top market cap quintile stocks' trades. It took a very long time because of the size of the data that I was working with and the inefficiency of my scripts-which I didn't detect and correct till Thursday.
Achievements for Wednesday
Same as on Tuesday.
Achievements for Thursday
Corrected the inefficiencies in my python script so that it finished in 2 hours rather than the estimated 30 days. Also, started counting the average number of algorithmic trades as a percentage of all trades every 5 minutes, on a monthly basis for the period 31 Oct 2006 to 26 Oct 2009. This is only for the top quintile stocks.