aeglon97 / evaluate-historical-stock-market-forecasts

Using the Bloomberg Terminal to Evaluate Stock Market Trends: An Analysis of Historical EPS Forecast Accuracy for All Firms in the 2019 S&P 500 Index

This project was conducted as my undergraduate senior thesis.

Motivation

For this project, I used Bloomberg Excel Functions to gather data on EPS and EOD prices for all firms in the 2019 S&P 500 Index. I assessed the quality of the data, cleaned and rearranged it, and stored the data in clean CSVs. I posed my own research questions, explored the data, and reported my findings through a final thesis report and Jupyter slides.

Introduction

This project was divided into 3 stages:

1. data wrangling
2. data exploration
3. data explanatory

The IPython Notebook files, data_cleaning.ipynb, data_exploratory.ipynb, and data_explanatory.ipynb all document the above 3 stages respectively. The first 2 files have been converted to HTML for convenience.

To view my Jupyter Notebook slides, clone the repository and render the data_explanatory.slides.html file in your browser.

Broad question: How do price forecasts for each firm in the S&P 2019 Index compare to their corresponding actual prices?

In order to address this broad question, I narrowed my focus down these factors: Forecasted Earnings-per-Share (EPS), Actual EPS, Actual End-of-Day (EOD) Price, and Forecasted EPS 3-months prior to that fiscal term. Since Earnings-per-Share directly determines a company's worth, I figured this would be a principle feature to use in measuring company value over the past 20 years. I received all of my data through the Bloomberg Terminals and their Excel functions, most especially the Bloomberg Help Desk.

Here is a breakdown of my 4 factors, their explanations, and how they contribute to the broad question:

• Forecasted EPS: this is will be subtracted from actual EPS to measure prediction errors of forecasts. This is also a central value to measure for the question of comparing historical forecasting accuracy of various firms. Forecasts are made based on a firm's fiscal period.

BDP("AAPL UW Equity","BEST_EPS","BEST_FPERIOD_OVERRIDE = 00Q1")

• Actual EPS: This is also measured based on fiscal period.

BDP("AAPL UW Equity","IS_EPS", "FUND_PER=Q1", "EQY_FUND_YEAR=2000")

• Actual EOD Price: the dates used for this measure correspond to the start of each calendar period, so it will be used to generate interesting visualizations and expand my analysis.

BDH("A UN Equity", "BEST_TARGET_PRICE", "03/31/1999", "03/31/2019")

• Forecasted EPS 3 months prior: EPS forecasts are made at the beginning of each fiscal period. This feature contains EPS forecasts made 3-months period to the current fiscal period. This will be an interesting metric to compare to forecasts at differing dates for the same firm.

BDH("AAPL US Equity", "BEST_EPS", "7/1/2007", "BEST_FPERIOD_OVERRIDE=08Q1", "days=a", "fill=p")

Initially, I intended to pick every firm present in the S&P 2019 Index for the past 20 years. However, the problem is that firms leave and join the S&P Index at unpredictable times, so analyzing EPS forecasts solely based on index alone instead of firms would be a highly misdirected approach.

Of course I could keep track of every firm that left and entered the S&P 500 index for the last 20 years. But due to the scope of this project, such an investigation was infeasible. I elaborate more on this under the Future Applications section.

It was more important to keep the same companies consistent across all datasets. To keep true to this approach, I picked the S&P 2019 firms as of November 2019.

Focus

For the rest of the project, I break down the broad question into these directions:

• compare both forecast and actual EPS for each firm in the S&P 2019.
• focus only and only on the firms present in the S&P 2019 Index.
• gather data projected from the past 20 years.

Data Overview

Here is a breakdown of the features among the final clean CSVs:

features.csv

• firm_id
• feature
• date
• term
• value

avgs.csv

• firm_id
• average
• average_type
• time_period
• feature

firms.csv

• firm_id
• firm

Research Questions

Question 1: Does average EPS prediction error depict any differences in trends whether by a yearly, quarterly, or full-term basis?

1. Forecasters were most optimistic in 2008Q4, and most pessimistic in 2000Q4.

2. The later the quarter, the more optimistic forecasters become in their average quarterly EPS predictions.

3. The year 2000 contains the highest variance among average prediction errors, ranging from -0.25 to -1.5.

4. The trends depicting EPS prediction error by quarter and year, separately, is consistent. All average EPS forecasts gather around 0 per year from 1999 - 2000, when ignoring outliers.

5. There was much more variance among quarterly EPS prediction errors than by a yearly or termly basis

Question 2: I generate naive EPS forecasts by calculating the rolling mean of the 2 actual EPS values from the past 2 quarters. How do my EPS forecasts compare to Bloomberg's EPS forecasts?

1. My naive forecasts tended to be overly optimistic in 2001Q1 and 2000Q4 (the latter being consistent with the Bloomberg forecasts). Meanwhile, my naive forecasts tended to be overly pessimistic in 2000Q4 and 2001Q1.

2. In the more recent years since 2012, Bloomberg EPS forecasts were less accurate than my own naive forecasts.

3. My naive forecasts spiked in pessimism during 2009Q1: only right before the Great Recession began to recover in June of 2009. Meanwhile, Bloomberg forecasters tended to be overly optimistic in their EPS forecasts during the "harder" years of the bear market.

4. Overall, my naive forecasts tend to be more accurate and better-fitting than Bloomberg forecasts across all years.

5. Across all quarters, my naive forecasts have always tended to be a better fit than the Bloomberg EPS forecasts.

Question 3: What differences and similarities emerge when analyzing the prediction error and percentage error of EPS forecasts?

1. Forecasters, on average, are likely to be more inaccurate in their predictions in Q4 of any given year.

2. For percentage errors, EPS forecasts percentage errors have become more inaccurate in the more recent terms, starting from 2014Q1.

3. The top 5 most inaccurate firm tickers for absolute prediction error are AGN, AIG, CHTR, LCRX, and VRSN. The most notable outlier is AIG, this is the only outlier on average for any given year.

4. The top 5 most inaccurate firm tickers for absolute prediction error are IBM, IRM, MCK, PXD, and QRVO. The most notable outlier is IRM, with a percentage error of over -1000 for EPS in the term 2014Q3.

Question 4: How do my naive RMSE values compare to Bloomberg RMSE when accounting for EPS forecasts? For what percentage of firms does my naive RMSE beat Bloomberg’s RMSE?

1. There are, in fact, more Bloomberg EPS RMSE values that are greater than my Naive EPS RMSE values per firm. This means that for the bulk majority of the data, my naive forecasts were much more reliable than the Bloomberg EPS forecasts.

2. Of all the RMSE values, 86.95% of my Naive EPS RMSE values are less than the Bloomberg EPS RMSE values.

3. After sorting each firm's Bloomberg and naive RMSE in decreasing order, the top 2 most inaccurately-predicted firm tickers are AIG and LRCX.

Question 5: How do EOD Prices trend across all firms from 1999 - 2019?

1. EOD prices show a general positive trend from 2009, as can be seen with figure (features-eod-term).

2. EOD prices see a "trough" from 2007 - 2009: the years of the Great Recession.

Limitations:

Overall

1. There was no explicit Bloomberg function to gather forecasted EOD price data by fiscal period for each firm in the S&P 2019.

2. There is an incongruence between fiscal periods and calendar periods. Forecasted and actual EPS are recorded by fiscal period, while EOD Prices and 3-month-priorly forecasted EPS are recorded by calendar period.

3. The dataset for the forecasted EPS 3 months prior was missing the year 1999.

eps_fc_terms

1. The year "1999" is missing data for all firms. This means when using this variable, we have to account only for the year 2000 onward.

eps_act

1. Contains 7 empty firms: BRK/B UN Equity, FOX UW Equity, GOOG UW Equity, HCP UN Equity, SYMC UW Equity, UA UN Equity

eps_fc_act

1. Has one firm with empty data: AMCR UN Equity

Future Applications

1. Look at other dates forecasted, not just 3 months before.
2. Implement linear regression models. Investigate multicollinearity to see if any other categorical or numerical value had any direct influence on affecting the generated p-values between forecasted and actual EPS.
3. Incorporate eps_fc_terms into the analysis. Look at other forecasts made from different periods, not just 3 months before. This would give me a greater insight into EPS forecasting trends made at various points in time before the current fiscal period.
4. Refocus my broad question to analyze variable relationships occurring during the Great Recession. Expand on how firms in the S&P 500 Index and their EPS and EOD prices fluctuate and responded to the stock market shifts in dynamic as a response to the economic recession.