Stock%20Value%20Ratio%20Classification

1
Stock Value Ratio Classification

• Yan Sui Zheng Chai

2
Classification

• MKV/BKV is an indicator of investors confidence
  in a particular company
• Being able to predict this ratio gives insight to
  predicting the stock prices

3
Outline

• Define Problem
• Data
• Method
• Initial Result
• Discussion

4
Definition

• Market Value
• The current quoted price at which investors buy
  or sell a share of common stock or a bond at a
  given time. Also known as "market price".
• Book Value
• The accounting value of a firm.
• The total value of the company's assets that
  shareholders would theoretically receive if a
  company were liquidated.
• Per share total value divided by number of shares

5
Problem Definition

• Given training data, predict the ratio for the
  future
• Classification vs Prediction Problem
• Need to define the classes (more later)

6
Problem Definition

• Why do we can about the ratio?
• Book value stays relatively constant and could be
  estimated
• Could estimate market price if we know this ratio
  and estimated book value

7
(No Transcript)
8
Outline

• Define Problem
• Data
• Method
• Initial Result
• Discussion

9
Data

• Dow Jones Industrial Average (Dow 30)
• Consists of 30 of the largest and most widely
  held public companies in the United States.
• E.g. American Express, ATT, Boeing, Citigroup,
  Exxon Mobil, GM, GE, Intel, etc.

10
Data

• wrds from Wharton
• Attributes are from CRSP/COMPUSTAT Merged
  database
• Book value and market value are from COMPUSTAT
  North America database
• High, low, and closing prices for each month are
  available

11
Problem

• Book value is updated annually
• 1 per year
• Market value is updated daily
• 365 per year
• What can we do?

12
Our Approach

• Estimate annual market price of a stock by
  averaging its high, low and closing prices over
  12 months.
• Market value estimated market price
• Another possibility
• Interpolate annual book values

13
Data Preprocessing

• Data Cleaning
• 400 attributes --gt 68 attributes (possibly
  more)
• Estimate annual market value
• Divide the MKV/BKV ratios into a number of
  classes
• Currently, there are 5 classes

14
1995 - 2005, 330 total observations
15
ratio class of Ex
R lt 2 0 20
2 lt R lt4 1 114
4 lt R lt6 2 97
6 lt R lt 10 3 69
R gt 10 4 30
16
Outline

• Define Problem
• Data
• Method
• Initial Result
• Discussion

17
Attributes

• Hundreds or even thousands possible attributes
• Using too many attributes may result in
  over-fitting
• Want to select a subset that work best for the
  task

18
Attribute Selection

• Select a subset of attributes to use
• Algorithms considered
• Greedy Algorithm
• Genetic Algorithm (genoud package in R)

19
Genetic Algorithm
20
Evaluation Function

• Produce a score of how a particular subset of
  features work (error rate)
• Minimization problem
• Possible candidates
• SVM
• Neural Network
• Etc.

21
Outline

• Define Problem
• Data
• Method
• Initial Result
• Discussion

22
Classify on the training data using 10 features
Error abs(predicted - actual)
23
Number of features
24
Top features
25
Explanation of Result

• Works well on training set
• When applied on new data, accuracy is around
  40-50

26
To Do List

• Retain more (non-atomic) attributes
• Try other evaluation functions
• Classification on daily ratio
• Other feature selection algorithms?
• Hopefully, find out which features are more
  influential in predicting market price for some
  stocks

27
Question?