BATT ASSIGNMENT:STOCK RELATIONS DEMYSTIFIEDBy Group 6
Mrinal Mantravadi
Gaurav Agarwal
Aseem Shandilya
OBJECTIVES OF THE STUDY Objectives:
To ascertain the relationship between the stock variations of IT industries (Here Infosys, TCS and Mahindra Satyam) , the indices of the BSE and the NYSE along with the GDP of USA
To determine the influence that the stocks of the IT companies have on each other
Reasoning: It is commonplace for the people to pass off the
lackadaisical performance of the IT companies with the US economy
Additionally, if one of the IT competitors is doing exceptionally well, it stands to reason that its competitors will face a slump
We wish to understand whether there is more to the performance of an IT company than just the variation of the US economy or the performance of a competitor
DESCRIPTION OF THE DATA Parameters
The parameters considered within the data are the following: Daily changes in the stock prices of three IT companies (Infosys, Mahindra
Satyam and TCS) Daily changes in the indices of the BSE Sensex and the NYSE: Euronext Quarterly data of the USA GDP for the years 2012-2013
Sources: The data for the stocks of the companies was sourced from Yahoo!
Finance The data for the indices of NYSE and BSE were sourced from their
respective websites The data for the quarter wise GDP of the USA was sourced from the
World bank
Treatments/Modifications It was observed that post 25/7/2013, Mahindra Satyam was known
as Tech Mahindra. Therefore, all the records were restricted till the above date
It was also observed that there was a time lag between the NYSE and the BSE The time lag was taken into account by lagging the NYSE data by 1 day as
opposed to the BSE
DATA ANALYSIS
All the data analysis done follows from this point on
COMPARISON OF VARIATIONS IN DATA-COMPANIES
Stocks from 17/12/2012 to 25/7/2013
Stocks from 17/12/2012 to 25/7/2013 Stocks from 17/12/2012 to 25/7/2013
Tata Consultancy Services Mahindra Satyam Infosys
The above data indicates a healthy amount of variation in the stocks of all three companies. At a glance, our data seems to be adequate for serving our purpose
COMPARISON OF VARIATIONS IN DATA-EXCHANGE INDICES
Stocks from 17/12/2012 to 25/7/2013
Stocks from 17/12/2012 to 25/7/2013
BSE Sensex NYSE: Euronext
At a glance, our sample seems to have captured a considerable amount of variation in the indices of both exchanges. As such, it should ensure that our analysis holds over the
population
STUDY OF CORRELATIONS IN DATA
To find out the relationship betweenIndian stocks, BSE and NYSE we started off with correlations.
In the adjoining graph we see significant positive correlation between TCS & INFY, TCS & BSE and MSAT & INFY.
However, we see a significant negative correlation between TCS and NYSE.
An important point to note here is that correlations only tell whether the variables are dependent on each other or not but give no idea on the causal dependence. Hence we decided to try determining causality
PANEL PROCEDURE – DESCRIPTION AND JUSTIFICATION Description
The PANEL procedure in SAS/ETS software fits classes of linear models that arise when time series and cross-sectional data are combined. It also tests if the data has any cross sectional effects/time series effects: random-effects and fixed-effects models: If there are no omitted variables – or if it is believed that omitted variables are uncorrelated with
the explanatory variables that are in the model – then a random effects model is to be used. It will produce unbiased estimates of the coefficients, use all the data available, and produce the smallest standard errors. However, it is possible that omitted variables will produce at least some bias in the estimates.
If there are omitted variables, and these variables are correlated with the variables in the model, then fixed effects models may provide a means for controlling for omitted variable bias. In a fixed-effects model, subjects serve as their own controls.
Justification/Adaptation We chose this method because our data is highly time sensitive and also cross sectional
Based on the above recommendations, we can expect our data to demonstrate fixed effects but not random effects as there are a lot of other factors we haven’t considered in our model, i.e. EPS, dividend policy, corporate governance policies etc. Therefore, our initial hypothesis would be that there are no random effects.
However, in keeping with the exploratory nature of our research, we have run tests on both fixed effects and random effects to test our initial hypothesis.
PANEL PROCEDURE ESTIMATES-INFOSYS(FIXED EFFECTS)Findings SAS Code Output
We run a text for fixed effects using proc panel with infy as the dependent variable and other stocks, indices and us GDP change as the predictor variables
proc panel data=stocks1;id qtr id;model infy = tcs msat bse nyse usgdpchg /FIXONE;run;
• The F test for no fixed effects has a null hypothesis that there are no fixed effects.
• Since we have a probability of 0.8, we fail to reject the null hypothesis.
• This clearly indicates that there are fixed effects and a random effects model would be unsuitable, however this hypothesis needs to be tested.
PANEL PROCEDURE ESTIMATES-INFOSYS(POOLED EFFECTS)Findings SAS Code Output
Here, we look at the pooled estimates, where the fixed and random effects are considered together. Even though the test for fixed effects proved positive, we cannot say for sure that
the random effects do not have some effect on the model.
The sas code is : proc panel data=stocks1;id qtr id; model infy = msat tcs bse nyse usgdpchg /POOLED;run;
• Overall model has an R sq. value of 0.08 which is expected because we have left out a large number of variables.
• However, there is a clear correlation between Msat and infy, showing that they exhibit pooled effects.
• The fact that the other variables seem to be insignificant cannot be taken at face value, therefore, we proceed to test for random effects.
PANEL PROCEDURE ESTIMATES-INFOSYS(RANDOM EFFECTS)Findings Comparison Output
• Identical coefficients predict a similar dependence and trend for Infosys stock’s interactions with the rest other variables.
• Test for random effects indicates random effects. Thus, we cannot rely on the test for fixed effects or pooled effects to rule out random effects.
• The model coefficients do not change, indicating that predictive power of these variables has been exhausted
PANEL PROCEDURE ESTIMATES-MAHINDRA SATYAM(FIXED EFFECTS)Findings SAS Code Output
It is also observed that the cross sectional effects have high probabilities indicating low significance
This is intuitive as the data seems to fit a time series model more appropriately than a cross sectional model
proc panel data=stocks1;id qtr id;model msat = tcs infy bse nyse usgdpchg /FIXONE;run;
• The F test for no fixed effects has a null hypothesis that there are no fixed effects.
• Since we have a probability of 0.45, we fail to reject the null hypothesis. This indicates that there are fixed effects and a random effects model might be unsuitable.
• However this hypothesis needs to be tested.
PANEL PROCEDURE ESTIMATES-MAHINDRA SATYAM(POOLED EFFECTS)Findings SAS Code Output
Here, we look at the pooled estimates, where the fixed and random effects are considered together. Even though the test for fixed effects proved positive, we cannot say for sure that
the random effects do not have some effect on the model.
The sas code is : proc panel data=stocks1;id qtr id; model msat = infy tcs bse nyse usgdpchg /POOLED;run;
• The R sq value observed in this model is also low (0.0773). Possible reasons:• Pooled effects considers the
combined effects of both the fixed and random effects. It does this by creating interaction variables. This might not provide a complete prediction for the dependent variable
• The model doesn’t include a number of factors e.g. EPS, dividend policy etc. as mentioned before
PANEL PROCEDURE ESTIMATES-MAHINDRA SATYAM(RANDOM EFFECTS)Findings Comparison Output
• Identical coefficients predict a similar dependence and trend for Mahindra Satyam’s stock’s interactions with the rest
• Test for random effects indicates random effects. Thus, we cannot rely on the test for fixed effects or pooled effects to rule out random effects.
• The model coefficients do not change, indicating that predictive power of these variables has been exhausted
PANEL PROCEDURE ESTIMATES-TCS(FIXED EFFECTS)Findings SAS Code Output
It is also observed that the cross sectional effects have high probabilities indicating low significance.
This is intuitive as the data seems to fit a time series model more appropriately than a cross sectional model.
proc panel data=stocks1;id qtr id;model tcs = msat infy bse nyse usgdpchg /FIXONE;run;
• The F test for no fixed effects has a null hypothesis that there are no fixed effects. • Since we have a probability
of 0.64, we fail to reject the null hypothesis. This clearly indicates that there are fixed effects and a random effects model might be unsuitable.
• However this hypothesis needs to be tested.
• Also, for tcs, the R sq. value is higher as compared to msat and infy which indicates that the data has a better fit for tcs as compared to infy and msat
PANEL PROCEDURE ESTIMATES-TCS(POOLED EFFECTS)Findings SAS Code Output
Here, we look at the pooled estimates, where the fixed and random effects are considered together. Even though the test for fixed effects proved positive, we cannot say for sure that
the random effects do not have some effect on the model.
The sas code is : proc panel data=stocks1;id qtr id; model tcs = infy msat bse nyse usgdpchg /POOLED;run;
• The R sq value observed in this model is low (0.1317), but much higher than the previous models (infy and tcs). Possible reasons:• Pooled effects considers the
combined effects of both the fixed and random effects. This might not provide a complete prediction for the dependent variable
• The model doesn’t include a number of factors e.g. EPS, dividend policy etc. as mentioned before
PANEL PROCEDURE ESTIMATES-TCS(RANDOM EFFECTS)Findings Comparison Output
• Similar coefficients indicate that both the models predict a similar dependence and trend in terms of Tcs stock’s interactions with the rest.
• Also, the test for random effects indicates that there are no random effects, as p= 0.23. This shows that we cannot rely on the test for fixed effects or pooled effects to rule out random effects.
• The model coefficients do not change, indicating that we have exhausted the predictive power of these variables.
MODEL COMPARISON AND VALIDITY CHECKSTata Consultancy Services Mahindra Satyam Infosys
The above graphs show that for all three stocks normality and heteroscedasticity is maintained.
Also, the models derived are a good fit
The change in the R sq. values is minimal but still shows that the best fit is obtained using the fixed effects model. Residuals are normal and showing different patterns for different quarters.
SUMMARY OF FINDINGSAND THE ROAD AHEAD
Summary of findings Fixed effects do exist in the model for all three companies Since pooled effects cover both random and fixed effects, it is difficult to isolate
the effect of either The results also indicate that the models drawn up thus far are a good fit and that
we are on the right track. However, the R-square values are low which indicates that there may be other
factors that need to be taken into account Our initial hypothesis that IT company stocks only depend on the US economy,
NYSE or competitors’ stocks is correct. However, these are not the only parameters
Road Ahead We can try incorporating parameters such as EPS, dividend policy etc. in the
model so as to have a better explanation of the variability of the IT stocks We can also attempt to cluster the IT company stocks by the type of clients they
cater to, the age of the company etc. to better understand the variation in stocks.
THANK YOU