Improving Price Momentum via News-Driven Returns Decomposition

September, 2019 / Issue 2 - Editorial on nontraditional data for Israeli Investors


In this post we summarize how intraday price momentum signal can be improved by utilizing granular low latency event news & sentiment feed and respective decomposition of stock returns into news-driven and non-news-driven ones

Overview


Market

  • Country/Region: North America, USA

  • Asset Class: Equities

  • Investment Universe: NYSE, NASDAQ, AMEX


Research & Application Context

  • Alt. Data Factor: News-driven price momentum

  • Method of Data Integration: Overlay

  • Underlying Inv. Strategy Type: Technical

  • Period of Analysis: 01/03/2000 - 31/10/2012

  • Investment Direction: Long-Short

  • Avg. Holding Period: week(s)

  • Avg. Portfolio Size: N/A

Investment Process

Underlying investment logic

In this case the underlying idea is a very short-term price momentum strategy - using daily returns and going L/S in best/worst 10% at the market close, holding for the next 5 days. Each day same procedure for 1/5th of the allocated capital is repeated (to avoid path dependency in the evaluation)

-> It is a direct baseline strategy to compare to.


Motivation

Some returns are just noise. Basic idea: consider returns only if relevant fundamental company-level news have been released during the same bar -> "news-driven" returns

Relevant, novel news feed overlay

  • Trading day is split into 15 min bars between 9:45 a.m. and 4:00 p.m. and the overnight bar between 4:00 p.m. on the previous trading day and 9:45 a.m.

  • For each stock fundamental news stream (s.a. M&A, analyst-ratings, credit, earnigns, product-services), collected and delivered by the data provider, is grouped into same 15 min bars

  • News feed filter - relevant and novel stories: only highly relevant news stories (event relevance score ERS =100), i.e. a story mentions predominantly a particular company, and with the event novelty score ENS =100, i.e. excluding repeated/republished news.

  • News-based returns decomposition: For each stock the 15 min bar retuns are then binary classified - news or no such news passing the filter during the given 15 min bar for a given stock

  • News-driven price momentum: Daily accumulation of only news-driven returns, ranking at the market close and going L/S in top/bottom 10% by the news-driven price momentum (one week holding period, each day for 1/5th of portfolio)

Main Results


1. Average, gross cumulative returns News-driven (blue) VS Non-news-driven (red) returns and their 95% confidence intervals over event time - every 30 minute until 4:00 p.m. on day t + 4. The average L/S difference in cumulative returns (winner-minus-loser portfolio) is plotted against the event day k, with the 95% confidence intervals.









2. Average, gross cumulative returns News-driven (blue) VS Non-news-driven (red) returns and their 95% confidence intervals over "event" day t+k after building a portfolio at day t..


--> Performance is persistent also over longer time horizons, no reversal for news-driven returns as in the regular price momentum





Investment Value of News Momentum

The table below reports descriptive statistics of daily returns of six investment strategies and their correlations:

  • the return of the news momentum strategy (NEWS);

  • the excess return on the market (MKT);

  • the average returns of non-market FFC4 factors: small-sized minus big-sized firms (SMB), high book-to-market ratio minus low book-to-market ratio firms (HML), high minus low prior 2 to 12 month return portfolios (UMD);the average return on the low prior month return portfolio minus that on the high prior month return portfolio (REV)

Return numbers are in monthly percentages (multiplying the percentage daily returns by 21); No transaction costs.



--> Given low/negative correlations, news-driven momentum factor can improve risk-adjusted returns when building a multi-factor strategy




Performance of the News Momentum Strategy


Remark: Results are without any transaction costs

  • Table (A): daily returns multiplied by 21 (for monthly approx.) and (monthly) alpha with respect to FFC4-factors

  • Table (B): betas with respect to FFC4 factors


Conclusion and Additional Remarks


Remark: interesting idea, BUT results presented in the draft-publication are only until 2012(!)

Performance in the most recent history not known


Decomposition of stock returns into news-driven and non-news-driven A more effective news-driven momentum strategy could be build by utilizing an intraday news feed to decompose stock returns into news-driven and non-news-driven when computing momentum.​

Stronger effect for smaller and less covered stocks Higher profitability of news-driven momentum strategy shown for smaller companies, lower analyst coverage and higher volatility - which comes however with higher costs to exploit this arbitrage (spreads, volume)


Persistence, no reversal in news-driven momentum News-driven returns are persistent in the short term and without reversals, as opposed to non-news-driven returns which precede a reversal.

News-driven momentum stronger after non-trading periods News-driven momentum on Monday (right after Friday & weekend) and right after opening is stronger than other weekdays and during the day, respectively, suggesting the "underreaction" of investors which drives the news momentum effect.

Source: "News Momentum", Hao Jiang, Sophia Zhengzi Li, Hao Wang - Rutgers Business School, Michigan State University, Prime Quantitative Research LLC, January 2018 (Draft)

DDQIR: Alternative Data-Driven Quantitative Investment Research

© 2020 by DDQIR