Extended Research of TRNA News Sentiment

IMPACT OF TRNA NEWS SENTIMENT ON PRICE

MOVEMENT

Tongli Zhang

Kimi Yang

IMPACT OF TRNA NEWS SENTIMENT ON PRICE

MOVEMENT

Tongli Zhang Kimi Yang

IMPACT OF TRNA NEWS SENTIMENT ON PRICE

MOVEMENT

Tongli Zhang Kimi Yang

IMPACT OF TRNA NEWS SENTIMENT ON PRICE

MOVEMENT

Tongli Zhang Kimi Yang

INDEX1. Data Structure and Preliminary Research

2. Introduction of Our Research Method about Relationship of News Sentiment and Daily Return

3. Analysis and Optimization of this Relationship

4. Trading Strategy Based on this Relationship

5. Analysis in Long Period of Time and on Intraday Basis

6. Rational behind the Empirical Relationship

DATA STRUCTURE AND PRELIMINARY RESEARCH

DATA STRUCTURE AND PRELIMINARY RESEARCH

DATA STRUCTURETRNA News Sentiment Database

Time of Record: Accuracy to ms

Relevance: From 0 to 1

Sentiment Score:

Positive, Neutral Negative, from 0 to 1

Time Period: Jan 2003- Nov 2011

Asset: Natural Gas, Coffee, Cotton etc.

Story type, Item general, etc.

94 background labels

Price Data

Daily Price Intraday Price

Origin Bloomberg Origin Pi-Trading

Time of Price

Date Time of Price

Minute by Minute

Asset Availableon Terminal

Asset Commodities,Stocks, FXs, ETFs

Time Period

Availableon Terminal

Time Period

VariousApril 2007 –August 2014 (Natural Gas)

Relation

IMPACT OF NEWS SENTIMENT ON PRICE

1. Aggregated Daily News Sentiment

Relevance weighted aggregated

2. Extreme News Sentiment on Daily Basis

The criteria for extreme news is the certain quantile of news sentiment score

For example (Extreme Positive News are news items with positive scoreshigher than 80% of the total news)

NATURAL GAS:NEXT DAY RETURN TO DAILY POSITIVE SENTIMENT VALUE

Correlation:0.0091Insignificant

NATURAL GAS:NEXT DAY RETURN TO DAILY NEGATIVE SENTIMENT VALUE

Correlation:0.016Insignificant

IMPACT OF SINGLE EXTREME NEWS

PosNeg

Neut

Integrated

Graph created by Wendi Zhu

ACCUMULATIVE IMPACT OF NEWS SENTIMENT &

RELATIONSHIP WITH PRICE

ACCUMULATIVE IMPACT OF NEWS SENTIMENT

& RELATIONSHIP WITH PRICE

ACCUMULATIVE IMPACT OF NEWS SENTIMENT

Overlap

•Daily Aggregated News Sentiment

Insignificant

•Single Extreme News Item

•All the Extreme News Items Published in One Day

Better?

ACCUMULATIVE IMPACT OF NEWS SENTIMENT

Number of Extreme News (Positive or Negative) and its relationship with price

1. Daily Basis: Daily Return Vs Daily Number of Extreme News

2. Use Year 2009 daily price data and sentiment data as training data set

3. Criteria for extreme news, sentiment score larger than 80% of news items

NEXT DAY RETURN VS NUMBER OF EXTREME POSITIVE NEWS

Correlation:-0.069SignificantSlope:-0.068

Num_days=250

NEXT DAY RETURN VS NUMBER OF EXTREME NEGATIVE NEWS

Correlation:-0.113SignificantSlope:-0.111

Num_days=250

OPTIMIZATION OF THE RELATIONSHIP

OPTIMIZATION OF THE RELATIONSHIP

OPTIMIZATION OF THE COEFFICIENT

1. Exclude abnormal points (Number>99%-tile)

2. Optimize criteria of extreme news(0.5~0.99)

3. Choose the prediction lag(0~5days)

EXCLUDE ABNORMAL POINTS

OPTIMIZE THRESHOLD FOR INCLUSION: POSITIVE

Best quantile:0.96Correlation: -0.20

OPTIMIZE QUANTILE: POSITIVE

Best quantile:0.94Correlation: -0.17

CHOOSE LAG

CHOOSE LAG

OPTIMIZED

1. Exclude abnormal days when number of extreme news larger than 99% of days

2. Criteria: Positive: sentiment score > 96% news items

Negative: sentiment score > 94% news items

3. Lag: One day

OPTIMIZED RESULT

Correlation:-0.200SignificantSlope:-0.192

Num_days=250

OPTIMIZED RESULT

Correlation:-0.173SignificantSlope:-0.167

Num_days=250

TRADING STRATEGY

TRADING STRATEGY

TRADING STRATEGY

Trigger of the Trade: Num_pos>11or Num_neg>9 or Num_pos+Num_neg>17

Type of Trading: short NG1 or United States Natural Gas ETF next day, positions settled daily

2003-01-022003-01-01 2003-01-042003-01-03

Next Return

Number of Extreme News Items

CUMULATIVE RETURN-UNG (2007/04 -2011/11)

0

50

100

150

200

250

300

10/10/2006 2/22/2008 7/6/2009 11/18/2010 4/1/2012 8/14/2013

Retu

rnCumulative Return of Three Strategies

Trading Strategy Buy&Hold Constantly Short

COMPARISON

Trading Strategy Buy & Hold Constantly Short

Cumulative Return

56.9% -67.9% 51.7%

Standarddeviation

20.2% 32.5% 47.5%

Max drawdown 20.9% 79.9% 32.6%

Sharp Ratio 0.614 -0.456 0.237

Calmar Ratio 0.594 -0.185 0.346

COST OF TRADING

Number of Trades: 235

-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

0.00% 0.20% 0.40% 0.60% 0.80% 1.00% 1.20%

Sharp

Rati

o

Transaction Cost Per Trade

Sharp Ratio-Transaction Cost

CUMULATIVE RETURN-NG1 (2003/01 -2011/11)

0

50

100

150

200

250

300

350

4/19/2001 9/1/2002 1/14/2004 5/28/2005 10/10/2006 2/22/2008 7/6/2009 11/18/2010 4/1/2012 8/14/2013

Retu

rn

Cumulative Return of Three Strategies

Trading Strategy Buy&Hold Constantly Short

COMPARISON

Trading Strategy Buy & Hold Constantly Short

Cumulative Return

92.6% -32.2% -94.1%

Standarddeviation

27.2% 45.3% 20.8%

Max drawdown 27.1% 63.7% 94.6%

Sharp Ratio 0.382 -0.080 -0.507

Calmar Ratio 0.383 -0.057 -0.112

Number of Trades: 370

INTRADAY DATA & RELATIONSHIP IN LONG

PERIOD OF TIME

INTRADAY DATA & RELATIONSHIP IN LONG

PERIOD OF TIME

RETURN VS NUMBER

Time point: 10:30, 11:30, 12:30, 13:30, 14:30, 15:30

Intraday Basis: Return of half hour after the time points Vs Number of Extreme News one hour before time point

10:309:30 12:3011:30

Number of Extreme News

Return

11:0010:00 12:00

RETURN VS NUMBER OF POSITIVE NEWS ITEMS

Insignificant

Num_items=1500

RETURN VS NUMBER OF NEGATIVE NEWS ITEMS

Insignificant

Num_items=1500

PosNeg

Neut

Integrated

Graph created by Wendi Zhu

INTRADAY PRICE MOVEMENT AROUND EXTREME NEWS ITEMS

A small number of extreme news happened in an hour, relatively low overlap

Price Movement

Extreme News Items

-30:00 30:00

NATURAL GAS90 QUANTILE POSITIVE

99.85

99.9

99.95

100

100.05

100.1

-40 -30 -20 -10 0 10 20 30 40

Retu

rn

Minutes After News Published

Return-Minute Number of Trades

127

Stand Deviation 0.47%

Max Return 1.07%

Min Return -1.56%

Positive Ratio 34.65%

Negative Ratio 57.48%

Zero Ratio 7.87%

P-value 6.84e-4

Return<0 Significant (99.9%)

NATURAL GAS90 QUANTILE NEGATIVE

99.95

99.97

99.99

100.01

100.03

100.05

100.07

100.09

-40 -30 -20 -10 0 10 20 30 40

Retu

rn

Minutes After News Published

Return-Minutes Number of Trades

165

Stand Deviation 0.74%

Max Return 4.97%

Min Return -2.22%

Positive Ratio 49.09%

Negative Ratio 44.24%

Zero Ratio 6.67%

P-value 1.37e-1

Return>0 86.3%significance

LONG PERIOD OF TIME

Regression between Monthly Return and Monthly Aggregated news sentiment

2003-02-012003-01-02 2003-04-012003-03-01

Aggregated News Sentiment

Next ReturnPre-Return Same-Return

LONG PERIOD OF TIME

Regression between Monthly Return and Monthly Aggregated news sentiment

Time of Return Correlation Significance

Positive Negative Positive Negative

Same Month 0.199 -0.263 Significant Significant

Previous Month 0.087 -0.113 Significant Significant

Next Month 0.034 -0.050 Insignificant

Insignificant

CONCLUSION & RATIONALE OF EMPIRICAL RESULTS

CONCLUSION & RATIONALE OF EMPIRICAL RESULTS

CONCLUSION

1. Aggregate news sentiment has general normal impact on return on a long time basis

2. A large portion of the news items may be about the past price movement, expectation, possible leak of information

Time of Return Correlation 95% Significance

Positive Negative Positive Negative

Same Month 0.199 -0.263 Significant Significant

Previous Month 0.087 -0.113 Significant Significant

Next Month 0.034 -0.050 Insignificant

Insignificant

CONCLUSION

1. Extreme News Items have a reverse impact on intraday price movement

2. Market overreaction and correction after

99.85

99.9

99.95

100

100.05

100.1

-40 -20 0 20 40

Retu

rn

Minutes After News Published

Return-Minute

99.95

99.97

99.99

100.01

100.03

100.05

100.07

100.09

-40 -20 0 20 40

Retu

rn

Minutes After News Published

Return-Minutes

CONCLUSION

1. Aggregated news sentiment does not affect daily return

2. Market overreaction and correction settled before next day

3. Number of extreme news items negatively impact next day return

4. Reflect the increase of uncertainty, increase of discount rate, thus lower price

Trading Strategy Buy & Hold Constantly Short

Cumulative Return

92.6% -32.2% -94.1%

Standarddeviation

27.2% 45.3% 20.8%

Max drawdown 27.1% 63.7% 94.6%

Sharp Ratio 0.382 -0.080 -0.507

Calmar Ratio 0.383 -0.057 -0.112

Thank You Q&A

TongliZhang/KimiYang,MasterofFinance,CareyBusinessSchool