The Influence of Extreme News - Xinwen Chen

1

Xinwen (Steven) [email protected]

Johns Hopkins University

2

Overview

11 Introduction & Data Preparation

33 Possible Trading Strategies & Performances • Buy & Hold (Benchmark)• Buy on Good News• Sell on Bad News

22 Method & Model Interpretation

3

Introduction & Data

TRNA Sentiment Data

Bloomberg Price Data

Merge…

TRNA:

•Thomson Reuters News Analytics

•News items are scored in real time during 11 years [ 2002 – 2012 ] for their relevance to underlying securities

Bloomberg:

•12 years [ 2003 – 2014 ] minute-by-minute price data during trading hours for each equity, commodity, and currencies

Introduction & Data >> Method & Model >> Trading Strategies & Performances

4

Introduction & Data

Concentrate on 30 Dow Jones Components (2003 ~ 2012):

TICKER NAME # of News Items TICKER

NAME # of News Items TICKER NAME # of News Items

MMM 3290 GE 14967 NKE 2090

AXP 3893 GS 9706 PFE 2963

AAPL 8884 HD 4026 PG 3487

BA 11738 INTC 7833 TRV 1697

CAT 3632 IBM 9876 UNH 3176

CVX 3567 JNJ 4576 UTX 2906

CSCO 7798 JPM 11674 VZ 15755

KO 2830 MCD 4398 V 2038

DD 3940 MRK 5852 WMT 2273

XOM 10036 MSFT 12631 DIS 6152

Introduction & Data >> Method & Model >> Trading Strategies & Performances

5

Can we use news items to predict the price changes

in the near future?

6

Method & Model

Logistic Regression Model:

where p: probability of binary-valued outcomes

X: predictor ( relevance × sentiment )

ln(p

1− p) = α + βX

Introduction & Data >> Method & Model >> Trading Strategies & Performances

Model#1: (predict good news)

p: probability of increase (1 = increase, 0 = otherwise)

X: relevance × positive sentiment

Model#2: (predict bad news)

p: probability of decrease (1 = decrease, 0 = otherwise)

X: relevance × negative sentiment

7

• Output of fitted model is the estimated probability of an price increase (decrease) following a given news item

• Need to find a probability threshold that leads to make a positive prediction (trade on the news)

Method & Model

Find the best cutoff value:

Example: (Boeing Company)

First 70% as training set

Last 30% as testing set

Validation Error Rate: 38.7%

Cutoff Value: 0.4801

Introduction & Data >> Method & Model >> Trading Strategies & Performances

8

Trading Strategies

Return

Benchmark: Buy & Hold• Open position at time 0 and close by the end of trading period

• take the profit / loss in the middle

Time t = 0

Time t = T

Introduction & Data >> Method & Model >> Trading Strategies & Performances

9

Trading Strategies

Strategy #1 : Buy on Good News• Open position only when the modeled probability above best

cutoff (good news)

• Hold till the next news item

Introduction & Data >> Method & Model >> Trading Strategies & Performances

Buy on Good News

Buy and Hold

TIME

10

Performance (Strategy #1)

Introduction & Data >> Method & Model >> Trading Strategies & Performances

11

Performances (Strategy #1)

Introduction & Data >> Method & Model >> Trading Strategies & Performances

TICKER NAME

AnnualOutperformance(vs. Buy & Hold)

TICKER NAME

AnnualOutperformance(vs. Buy & Hold)

TICKER NAME

AnnualOutperformance(vs. Buy & Hold)

BA 13.82% CAT 11.80% CVX 11.35%

CSCO 10.38% KO 7.86% NKE 6.53%

PG 5.82% DD 5.56% PFE 4.70%

GS 4.65% XOM 2.52% UNH 2.24%

MMM 1.58% V 1.13% MCD 0.98%

JPM 0.52% JNJ 0.20% INTC - 3.55%

MSFT - 5.45% UTX 5.52% AXP - 7.35%

TRV - 9.00% DIS - 9.52% WMT - 11.44%

GE - 13.63% MRK - 14.44% VZ - 19.61%

IBM - 20.62% HD - 30.12% AAPL - 37.32%

12

Trading Strategies

Strategy #2 : Sell on Bad News• Open position at time 0 and hold till the bad news occurs

• Re-open position after 2 hours

Introduction & Data >> Method & Model >> Trading Strategies & Performances

Buy and Hold

TIME

2 Hours

Sell on Bad News

13

Trading Strategies

Strategy #2 : Sell on Bad News• Open position at time 0 and hold till the bad news occurs

• Re-open position after 2 hours

Introduction & Data >> Method & Model >> Trading Strategies & Performances

Buy and Hold

TIME

2 Hours

Sell on Bad News

14

Performances (Strategy #2)

Introduction & Data >> Method & Model >> Trading Strategies & Performances

15

Performances (Strategy #2)

Introduction & Data >> Method & Model >> Trading Strategies & Performances

TICKER NAME

AnnualOutperformance(vs. Buy & Hold)

TICKER NAME

AnnualOutperformance(vs. Buy & Hold)

TICKER NAME

AnnualOutperformance(vs. Buy & Hold)

CAT 13.32% MMM 10.40% GS 6.24%

XOM 5.34% GE 4.15% PG 3.77%

KO 3.74% MSFT 2.60% JPM 1.77%

CVX 1.58% INTC 1.57% BA 1.51%

AAPL 0.89% UTX 0.57% WMT 0.53%

HD 0.28% CSCO 0.17% VZ 0.12%

MRK 0.11% MCD 0.11% IBM 0.04%

PFE 0.04% UNH 0.02% V - 0.01%

TRV - 0.01% AXP - 0.28% DD - 0.86%

DIS - 1.57% NKE - 4.24% JNJ - 5.49%

16

Conclusions

• We are able to identify several stocks that appear to be quite sensitive to news.

• Both “buy on good news” and “sell on bad news” strategies outperform the benchmark for majority stocks.

• For “buy on good news” strategy, the number of stocks for which we achieve out-performance is not significantly different from 50% of them

• For “sell on bad news” strategy, 50% stocks only have a small out-performance comparing with the benchmark.

17

Q & A