FINANCIAL MARKET: Assignment 3
14 octobre 2013

Alberto de Arriba Meriem Berrada Master of Finance Universidad Carlos III de Madrid

Marcos Lain Lianna Petrosyan Scolar Year 2013-2014

1 INTRODUCTION

1

Introduction

The main aim of this third assignment is to compare between the evolution of spread and depth using two dierent ways to assemble stocks in portfolios.

1.1

Data

We have one month of data concerning 40 rms quoted in stock Exchange per minute. For each stock, for each day, we have a detailed amount of messages, detph for ask and bid and the volume traded. AA ADBE AMAT AMZN BHI BRCM CELG COST CTSH DIS EBAY GE GILD GOOG HON INTC KMB MMM PFE PNC
1.2 Instructions

Alcoa Adobe System Applied Material Amazone.com Baker Hughes Inc Broadcom Celgene Costco Wholesale Cognizant Technology Walt Disney Ebay General Electric Gilead Sciences, Inc Google Honeywell International Intel Corp Kimberly-Clark 3M Corp Pzer Pnc Finance

AAPL AGN AMGN AXP BIIB CB CMCSA CSCO DELL DOW ESRX GENDZ GLW GPS HPQ ISRB KR MOS PG SWIN

Apple Inc Allergan Inc Amgen Inc American Express Co Biogen Idec Inc Chubb Comcast Cisco Dell Dow Chemical Express Scripts Holding Corning inc Gap inc Hewlett-Packard intuitive surgical Kroger Mosaic Corp Procter and Gamble Southwestern Energy

 Construct 4 equally portfolios with the average of message by stock as criteria.  Plot the graphs reprensenting the spread, depths, messages and volume for these portfolios.  Construct 4 equally portfolios with the average of volume by stock as criteria.  Plot the graphs reprensenting the spread, depths, messages and volumes for these new portfolios.  Compare between the 2 constructions of the portfolios depending on the criteria.

2

2 METHODOLOGY

2

Methodology
1. Filter the data by minut, so as we can have all the information concerning the variable (spread, message, volume and depth) 2. Compute the average of all these variables for each stock 3. Sort the averages according to the criteria (messages or volume) ascending 4. Construct 4 portfolios equally weighted : 10 rst ⇒ portfolio "Low" 10 second ⇒ portfolio "Medium Low" 10 third ⇒ portfolio "Medium High" 10 fourth ⇒ portfolio "High" 5. For this minut, comput the average per variable per portfolio

As this process should be repeated a lot of times, we have coded a macro in VBA, to compute the averages, construct the portfolio and compute the second averages. The user has to select the minut by applying the lter in the column "H", then copy the selection in the sheet "data2". Afterwards, he has to press one of the button "per message" or "per volume" so that to select the criteria with what he wants to construct the portfolio.

By executing the macro, the result is printed in the sheet "result". The only problem is that this macro can't save the nal results, so in order to have all the averages for all the minuts, we have to copy this result in another sheet. Finally, to plot the graphs, we only have to select the columns that interest us.

3

3 RESULTS

3
3.1

Results
Constitution of portfolios

Constitution of the portfolio with message

Figure 1  Evolution of the average of messages by stock
Constitution of the portfolio with volume

Figure 2  Evolution of the average of volume by stock

4

3 RESULTS

As we have explained before, we have 4 equally weighted portfolios in each section and the results can be appreciated in the following graphs. We have noticed that the constitution of the portfolios doesn't change a lot during the day. These graphs have been ploted with the data of 936 minute. As we can see the majority of the most important companies in our sample remain in the rst portfolio regardless of the use of the criteria. This is for instance the case of Apple, Dell, Amazon or Cisco, all of them related with the technological world. It seems to be happen the same as we are analyzing the lowest companies ; they remain at the bottom of our graphs. This fact could give strong evidence that the number of messages received per minute and the volume traded in the transactions are positively related with the size of the company. Finally we can see that the gap between the top rms and the rest ones are very prominent, especially the rst 5 rms in the sample.

5

3 RESULTS

3.2

Evolution of the average of messages

Figure 3  Evolution of the average of messages for the portfolios constructed with message

Figure 4  Evolution of the average of messages for the portfolios constructed with volume
In this section we can observe that the largest amount of messages is received in the rst minutes of the day (more than 10000 messages in the rst portfolio) in order to fall considerably several minutes later. During the day we appreciate how the fall it is moderate and it experiments a slight rise at the end of the session. The pattern seems to be similar if we observe the portfolios built by messages and the portfolios built by volume ; nevertheless we could say that the gap between the top rms and the rest is more prominent with the use of the rst criteria (messages).

6

3 RESULTS

3.3

Evolution of the average of spread

Figure 5  Evolution of the average of spread for the portfolios constructed with message

Figure 6  Evolution of the average of spread for the portfolios constructed with volume
These graphs tested empirically that (as we displayed in the previous paper) at the beginning of the session the spread is larger than at the rest of the day, so the liquidity is lower. We can also appreciate that the companies who belong to the third and four portfolios have larger levels of spread (at least at the beginning of the session) than the top rms, and this fact happen regardless of the criteria. It is also true that except these rst moments, the average of the spread remains almost constant the rest of the day and we cannot observe main dierences between the lowest and the largest companies. If we analyze both criteria, we observe that the gap is a little bit larger if we use the volume as criteria than if we use the number of messages. 7

3 RESULTS

3.4

Evolution of the average of depth

Figure 7  Evolution of the average of depth for the portfolios constructed with message

Figure 8  Evolution of the average of depth for the portfolios constructed with volume
The depth in our sample has a positive relationship with the pass of the time. That means that as the day is advancing, the depth increases regardless of the portfolios. It shouldn`t be a surprise that the larger amount of depth is achieved by the top rms and this fact happen in both plots. Despite of this, we can appreciate that the gap between the top rms and the rest increases in a more prominent way if we use the volume as criteria than if we use the number of messages (that could be caused by the decline of the level of the second group). 1
1. The depth which is plotted is the depth for ask.

8

3 RESULTS

3.5

Evolution of the average of volume

Figure 9  Evolution of the average of volume for Portfolio with message

Figure 10  Evolution of the average of volume for Portfolio with Volume
The previous graphs give us strong evidence that the largest levels of volume are reached at the beginning and at the end of the session. That means that the most important transactions are carried out at these particular periods. The rms of the rst group practically perform the most important actions and that is why the gap is so prominent respect to the rest. If we use the volume criteria we can observe that the dierence in volume transactions is larger than if we had used the messages as criteria. This is normal because we have built the portfolios based on the amount of volume of each company. 9

4 CONCLUSION

4

Conclusion

We have tested that the composition of the portfolios is quite similar in both cases (volume criteria and messages criteria) because these two features are positively related. We think that the size of the companies is the main feature that allows to the rm to be placed in the rst or in the second portfolio. We have displayed that the patterns of spread, number of messages, depth and volume are very similar regardless of the method that we have used. Despite of this fact, we can observe dierences in gap between the top rms and the rest depending on the variable and depending on the method. 2

2. The X axis label should be hours and no gures from 1 to 1300 it is due to the fact that the funtion to plot graph didn't recognize that the data was "hours"

10