Fama-French Three-Factor ETF Model

Introduction

The purpose of this article is to propose an implementation of the Fama-French Three-Factor model using sentiment analysis data on ETFs. Exchange Traded Funds (ETFs) are marketable securities that are traded on the stock exchanges daily. There are multiple types of ETFs, each looking to accomplish different objectives such as tracking an index. Sector based ETFs look to give investors exposure to certain sectors and industries by allowing for a purchase of a basket of assets through the ETF.

This report will look towards building a 3-Factor model using historical returns, volatility, and sentiment to explain the future returns of ETFs. The model will then be used to select the best ETFs for a trading strategy.

Model

The following 3-Factor model will be considered in this report:

Fama-French Three-Factor ETF Sentiment model

This model aims to predict the future performance of an asset using its past performance, past volatility, and past sentiment.

US Sectors

Firstly, we will attempt to test the model on US Industry/Sectors ETFs.

The following Sector ETFs will be used:

  1. Vanguard Energy ETF(VDE)
  2. Vanguard Financials ETF(VFH)
  3. Vanguard Health Care ETF(VHT)
  4. Vanguard Industrials ETF(VIS)
  5. Vanguard Information Technology ETF(VGT)
  6. Vanguard Materials ETF(VAW)
  7. Vanguard Telecommunication Services ETF(VOX)
  8. Vanguard Utilities ETF(VPU)
  9. They represent 8 unique sectors in the US Market.

After fitting the model, at the start of each month, we will calculate the predicted future performance of all sectors and then long the ETF of the sector with a largest predicted return and hold for 1 month before we repeat the process.

For the sentiment, stocks from the S&P500 index are grouped into sectors and the daily sentiment of the stocks in each sector group is averaged to obtain the daily sentiment of each sector.

For the return and volatility data for each sector, price data from Vanguard Sectors ETFs are used as a proxy.

The Fama-French Three-Factor model is then fitted using linear regression and the following model coefficients are obtained:

Fama-French Three-Factor ETF Sentiment model 2

Interestingly, the negative coefficient for 3 Month Historical Return implies that better past performance leads to poorer future performance. A possible reason is that investors buy into sectors that have performed well in the past leading to the sector becoming overvalued and perform poorly in subsequent months.

Also, a larger volatility which is associated with more risk leads to better future performance which is somewhat consistent with the Capital Asset Pricing Model.

Lastly, better sentiment obviously leads to better performance and poor sentiment lead to poorer performance.

Here are the backtest results from Jan 2016 to June 2017:

Fama-French Three-Factor ETF Sentiment model 3 - Backtest

**Benchmark is the S&P500 Index

There is some merit to selecting individual sectors over investing in the broad index, leading to larger returns over time.

European Indexes

The returns and sentiment of the following European Indexes will be used in this section:

  1. AEX
  2. CAC
  3. DAX
  4. IBEX
  5. STOXX50

The Fama-French Three-Factor model is fitted using linear regression with the following results:

Fama-French Three-Factor ETF Sentiment model Eur 1

Again, we find that there is a negative coefficient for 3 Month Historical Return implying that better past performance leads to poorer future performance.

However, the significance of the volatility factor appears small and could possibly be eliminated to form the reduced model below.

Fama-French Three-Factor ETF Sentiment model Eur 4

Fitting the reduced model using the same data produced the following results:

Fama-French Three-Factor ETF Sentiment model Eur 2

Using the same strategy of investing in the ETF with the highest predicted return for 1 month, the following results were obtained[1]:

Fama-French Three-Factor ETF Sentiment model Eur 3

[1] These results were obtained from trading the indexes. Index ETFs should replicate the returns of the indexes and lead to similar performance barring ETF discounts and premiums. Commissions and slippage is simulated at 0.1% of trade value. The Euro Stoxx 50 index is used as the benchmark index. Risk Free Rate of 1% per annum is used for calculation of Alpha, Beta, Sharpe and Sortino.

The performance of the model is extremely close to that of the benchmark. While the benchmark, Euro Stoxx 50 is one of the indexes being considered by the model, during the 1 year backtest period, the model did not select and invest in the Euro Stoxx 50 index.

This is evidence of the strong correlation between the countries that reside in the Eurozone. However, the model did however outperform the index slightly despite being handicapped by transaction costs which shows that while there is strong correlation, the model is able to pick out the better performers.

Asian Indexes

The returns and sentiment of the following Asian Indexes will be used for this report:

  1. Straits Times Index
  2. BSE SENSEX
  3. Shanghai Composite
  4. Nikkei
  5. Taiwan Stock Exchange Weighted Index
  6. KOSPI
  7. Jakarta Composite

The Fama-French Three-Factor model is fitted using linear regression with the following results:

Fama-French Three-Factor ETF Sentiment model Asia 1

Again, we find that there is a negative coefficient for 3 Month Historical Return implying that better past performance leads to poorer future performance.

Using the same strategy of investing in the ETF with the highest predicted return for 1 month, the following results were obtained[2]:

Fama-French Three-Factor ETF Sentiment model Asia 2

[2] These results were obtained from trading the indexes. Index ETFs should replicate the returns of the indexes barring ETF discounts and premiums. Commissions and slippage is simulated at 0.1% of trade value. Straits Times Index is used as the benchmark index. Risk Free Rate of 1% per annum is used for calculation of Alpha, Beta, Sharpe and Sortino Ratios. Constant exchange rate is assumed.

Conclusion

The use of factor models to predict future returns can allow for the selection of ETFs with superior performance, leading to outperformance of benchmarks.

More about the Data

The data used in this report is powered by FinSentS published by InfoTrie Financial Solutions.

The FinSentS News Sentiment database offers daily media sentiment indicators for 23,000+ global equities, calculated by applying sophisticated real-time machine-learning algorithms to the content of thousands of news websites and media sources from around the world.

Each stock has 5 indicators:
– Sentiment Score: a numeric measure of the bullishness / bearishness of news coverage of the stock.
– Sentiment High / Low: highest and lowest intra-day sentiment scores.
– News Volume: the absolute number of news articles covering the stock.
– News Buzz: a numeric measure of the change in coverage volume for the stock.

News analytics | Alternative data

Send us a mail : contact@infotrie.com

Check our market blog or our AI and NLP blog

Want to access our data? Try our API on api.infotrie.com or access our FinSentS platform.


itrie

InfoTrie Admin

0 Comments

Leave a Reply

Your email address will not be published. Required fields are marked *