Strathclyde Business School, finTech MSc

As part of the fall 2018 iteration of the ‘becoming an effective technoloy analyst’ class of the Strathclyde Business School finTech MSc program coursework, below are the instructions for your data-science finance assignment. Examples solutions in both the R and Python programming languages will be provided in due time.

Full stack data-science finance (small) project

Preprocessing (ELT)

Extract

Minimum required

In an excel woorkbook, query Bloomberg for historical (bdh) as well as contemporaneous (bdp) data for a market index as well as a broad cross-section of U.S. stocks. Historical data should be retrieved from October 1st 2016 to today at the daily frequency on individual ticker specific sheets (one sheet per name). All names’ contemporaneous data, on the other hand, should sit on a single sheet. The Bloomberg ticker for the market index is ‘RAY Index’ while those for the corporation names are listed below:

BBG stock tickers
ADM US Equity CIVI US Equity GBX US Equity LIND US Equity SERV US Equity
AE US Equity CLGX US Equity GDI US Equity LZB US Equity SGA US Equity
AGCO US Equity CLR US Equity GHC US Equity MAN US Equity SITE US Equity
AJRD US Equity COMM US Equity GME US Equity MEI US Equity SMP US Equity
ALG US Equity CRL US Equity GOLF US Equity MLR US Equity SPXC US Equity
AMD US Equity CTB US Equity GPN US Equity MRC US Equity STRT US Equity
AMOT US Equity CTLT US Equity GTLS US Equity MTD US Equity SUPN US Equity
ASGN US Equity CTXS US Equity HFC US Equity MTZ US Equity TAST US Equity
ATRO US Equity DHI US Equity HOFT US Equity NC US Equity TMO US Equity
AVT US Equity DKS US Equity HPE US Equity NGVT US Equity TNET US Equity
AWI US Equity EBIX US Equity HURC US Equity NHC US Equity TPB US Equity
BBBY US Equity EEFT US Equity HWKN US Equity NUE US Equity UBNT US Equity
BFAM US Equity ELF US Equity HY US Equity OSIS US Equity UFPI US Equity
BID US Equity ELVT US Equity IAC US Equity OSK US Equity UFS US Equity
BIG US Equity EML US Equity IART US Equity PFGC US Equity USAK US Equity
BKNG US Equity ENTG US Equity IBP US Equity PGTI US Equity VLGEA US Equity
BLD US Equity ERI US Equity IDTI US Equity PKI US Equity VLO US Equity
BSET US Equity ETH US Equity INT US Equity PLPC US Equity VRSK US Equity
BWA US Equity FICO US Equity IOSP US Equity PRAH US Equity WBC US Equity
BYD US Equity FISV US Equity ITRI US Equity PSX US Equity WERN US Equity
CAL US Equity FL US Equity JLL US Equity RBC US Equity WGO US Equity
CBRE US Equity FLR US Equity KHC US Equity RS US Equity WRK US Equity
CENTA US Equity FLT US Equity KSU US Equity RXN US Equity XPO US Equity
CHEF US Equity FTV US Equity LGND US Equity SCL US Equity ZBRA US Equity

The historical time series should include the following market & book data fields:

Field Bloomberg symbol
close price PX_LAST
book value per share BOOK_VAL_PER_SH
earnings per share TRAIL_12M_EPS
dividend per share TRAIL_12M_DVD_PER_SH
debt SHORT_AND_LONG_TERM_DEBT
equity TOTAL_EQUITY
current assets BS_CUR_ASSET_REPORT
current liabilities BS_CUR_LIAB
sales SALES_REV_TURN

Contemporaneous data on the other hand should include the number of shares outstanding, number of directors on the board, number of women on the board, number of board meetings per year, long company name and company description. Explore Bloomberg to find the corresponding field symbols.

Going further

  • Using VBA, make your workbook updatable. Ammend your workbook so that it retrieves up to date data in one clic. I.e. if in the future you open the workbook you created today, the workbook should be able to retrieve up to date data.
    • Hint 1. Update doesn’t necessarily mean adding most recent values to an existing time series. Requerying the whole data up to the most recent date would work as well.
    • Hint 2. Inspect the BQL syntax in Bloomberg formula cells, ammend accordingly.
  • Using VBA, make your workbook flexible. Ammend your workbook so that it can retrieve data for any set of stocks/indexes & market/book fields at various frequencies (year, month, week, day), from and to any date. The user should only have to list the tickers/fields and set the parameters on one sheet.
    • Hint 1. Object oriented programming could help; excel table objects in particular.
    • Hint 2. Create an ‘update’ sheet with tickers list, parameters (frequency, start and end dates) and fields. This sheet could also be used to host the contemporaneous dataset.
  • Using VBA, make your workbook fully portable. If you open your workbook without a live Bloomberg connection you’ll notice you loose the contemporaneous dataset; try to fix that problem somehow.
    • Hint 1. VBA events could help.

You now have a fully portable, customizable Bloomberg financial data extraction tool and now it’s time to use it.

Load

Using R or Python (example solutions will be provided for both programming languages), load the workbook data in memory. Organise the data in two dataframes, one for the historical times series, the other for static (contemporaneous) data. The time series dataframe should have a two-level row index including tickers & dates while columns should host the corresponding time series; the dataframe should broadly look like this:

##                ticker       Date  PX_LAST BOOK_VAL_PER_SH TRAIL_12M_EPS
##     1:      RAY Index 2016-10-04 1273.897        476.0300       58.4400
##     2:      RAY Index 2016-10-05 1279.598        476.0500       58.4400
##     3:      RAY Index 2016-10-06 1279.430        476.0500       58.4300
##     4:      RAY Index 2016-10-07 1274.602        476.0700       58.4800
##     5:      RAY Index 2016-10-10 1281.312        476.0800       58.4700
##    ---                                                                 
## 62692: USAK US Equity 2018-10-17   17.830          8.4359        0.7559
## 62693: USAK US Equity 2018-10-18   17.310          8.4359        0.7559
## 62694: USAK US Equity 2018-10-19   17.410          8.4359        0.7559
## 62695: USAK US Equity 2018-10-22   18.240          8.4359        0.7559
## 62696: USAK US Equity 2018-10-23   17.310          8.4359        0.7559
##        TRAIL_12M_DVD_PER_SH SHORT_AND_LONG_TERM_DEBT TOTAL_EQUITY
##     1:                   NA                  566.360      504.630
##     2:                   NA                  566.370      504.650
##     3:                   NA                  566.340      504.640
##     4:                   NA                  566.360      504.660
##     5:                   NA                  566.350      504.660
##    ---                                                           
## 62692:                    0                   88.958       70.125
## 62693:                    0                   88.958       70.125
## 62694:                    0                   88.958       70.125
## 62695:                    0                   88.958       70.125
## 62696:                    0                   88.958       70.125
##        BS_CUR_ASSET_REPORT BS_CUR_LIAB SALES_REV_TURN
##     1:             339.210     234.730             NA
##     2:             339.340     234.820             NA
##     3:             339.340     234.820             NA
##     4:             339.580     234.980             NA
##     5:             339.610     235.000             NA
##    ---                                               
## 62692:              78.798      71.077        135.381
## 62693:              78.798      71.077        135.381
## 62694:              78.798      71.077        135.381
## 62695:              78.798      71.077        135.381
## 62696:              78.798      71.077        135.381

The static dataset on the other hand should be row-indexed by tickers and have columns hosting the corresponding static data fields. For static data, only numeric fields should be loaded with long company name and description fields left to the excel workbook for reference.

Transform

Market betas

Minimum required

Using the most recent samples in the time series data, calculate the individual 1-year market betas for the stocks. Show calculations and comment. Comments should include a detailled discussion on what market betas are, what they represent for stocks as well as details about the corresponding model. Plot your results as a histogram and comment. Hint: there are 252 trading days in a year.

Going further

Using all the time series samples, calculate the individual rolling 1-year market betas for the stocks. Randomly select five stocks and display their corresponding rolling beta time series on the same lineplot.

Features interactions

  • Using the most recent samples in the time series dataset, for each name construct a set of feature interactions that include the following popular financial ratios: price to book, price to earnings, dividend yield and gearing. Show calculations and discuss these concepts from a corporate finance standpoint.
  • Explore this new dataset. Hint: use visualization tools.

Modeling

Minimum required

Cluster analysis (unsupervised learning)

Hierarchical clustering

After normalizing the ratios dataset above to zero means and unit variances, apply hierachical clustering and draw the corresponding dendogram. What seems to be the optimal number of clusters for this dataset? Explain.

K-means
  • Implement a two-cluster k-means analysis on this dataset. Explore the resulting cluster characteristics: calculate the cluster specific means for each ratio. Comment on the results and propose labels for the two classes. Hint: how would Warren Buffett most likely answer this?
  • Label individual names accordingly in a new ‘classes’ dataframe.

Going further

Classification (supervised learning)

  • Create a betas dataset that subsets the most recent (last sample date) samples from the rolling maket betas dataset above. Merge the classes, ratios, static and betas datasets together.
  • Implement a classification analysis on the resulting dataset where the target is name’s class as attributed above. Use various classifiers including logistic regression, k-nearest-neighbours, support vector machines, decision tree, random forest and neural network (multi-layer perceptrons). Use 75-25% for training-test sets split and 5-fold cross-validation.
  • For each model:
    • Show training and test set confusion matrices and calculate corresponding precision & recall indicators; comment. Your comments should include a discussion on precision and recall.
    • Explain what the model does and how. Discuss model paramaters and how they contribute to model fine-tunning.
    • Find model optimal parameters using gridsearch and run model accordingly. Show corresponding learning curves.