Quarterly Mark December 2023 - 19

EXHIBIT 9
Out-of-Sample Cumulative Log Returns of Portfolio Sorted Based on Sentiment Score Constructed from News Articles
15
-
L
10
-
L-S
S
L-S
L
S
5
SPY
EW
EW
EW
VW
VW
VW
-5
-10
2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020
NOTES: The sentiment scores are developed based on a dictionary-based statistical method. L: long only, S: short only, L-S: long-short,
EW: equally weighted, VW: value weighted, SPY: S&P500. Long portfolio is constructed by buying 50 stocks with maximum positive
sentiment, and short portfolio is constructed by shorting 50 stocks with the most negative sentiment. The equal weighted long-short
portfolio achieves an annualized Sharpe ratio of 4.21 while value weighted long-short portfolio achieves a Sharpe of 1.24. The portfolios
obtained from the sentiment scores outperform the baseline of the S&P 500.
SOURCE: Ke et al. 2019, Figure 5©.
relevant information from news articles, such as company news, earnings releases,
and analyst reports. This information can be used in the investment decisionmaking
process and improve portfolio management. Earnings call analysis involves
using NLP to process and analyze earnings call transcripts. The goal is to extract
key financial metrics and sentiment indicators to better understand a company's
financial performance when making investment decisions.
Before the adoption of deep learning, a variety of NLP methods were widely
used in stock price prediction, including keyword spotting (Wuthrich et al. 1998),
bag-of-words classification (Tetlock, Saar-Tsechansky, and Macskassy 2008),
manually crafted lexicons (Das and Chen 2007), pre-existing lexicons (Azar and
Lo 2016), mixture models (Si et al. 2013), and lexicon-based statistical modeling
(Ke et al. 2019). They were based on data obtained from professional periodicals,
aggregated news, Twitter, financial reports, and so on. Exhibit 9 shows the
performance of a long-short portfolio (inter-day trading) constructed using
sentiment score from Ke et al. (2019). They found that portfolios constructed using
dictionary-based statistical models can obtain an annualized Sharpe ratio of 4.21,
implying NLP-based sentiment scores can be used for inter-day trading. Similarly,
Quarterly Mark | 19

Quarterly Mark December 2023

Table of Contents for the Digital Edition of Quarterly Mark December 2023

Quarterly Mark December 2023 - Cover
Quarterly Mark December 2023 - Inside Front Cover
Quarterly Mark December 2023 - 1
Quarterly Mark December 2023 - 2
Quarterly Mark December 2023 - 3
Quarterly Mark December 2023 - 4
Quarterly Mark December 2023 - 5
Quarterly Mark December 2023 - 6
Quarterly Mark December 2023 - 7
Quarterly Mark December 2023 - 8
Quarterly Mark December 2023 - 9
Quarterly Mark December 2023 - 10
Quarterly Mark December 2023 - 11
Quarterly Mark December 2023 - 12
Quarterly Mark December 2023 - 13
Quarterly Mark December 2023 - 14
Quarterly Mark December 2023 - 15
Quarterly Mark December 2023 - 16
Quarterly Mark December 2023 - 17
Quarterly Mark December 2023 - 18
Quarterly Mark December 2023 - 19
Quarterly Mark December 2023 - 20
Quarterly Mark December 2023 - 21
Quarterly Mark December 2023 - 22
Quarterly Mark December 2023 - 23
Quarterly Mark December 2023 - 24
Quarterly Mark December 2023 - 25
Quarterly Mark December 2023 - 26
Quarterly Mark December 2023 - 27
Quarterly Mark December 2023 - 28
Quarterly Mark December 2023 - 29
Quarterly Mark December 2023 - 30
Quarterly Mark December 2023 - 31
Quarterly Mark December 2023 - 32
Quarterly Mark December 2023 - 33
Quarterly Mark December 2023 - 34
Quarterly Mark December 2023 - 35
Quarterly Mark December 2023 - 36
Quarterly Mark December 2023 - 37
Quarterly Mark December 2023 - 38
Quarterly Mark December 2023 - 39
Quarterly Mark December 2023 - 40
Quarterly Mark December 2023 - 41
Quarterly Mark December 2023 - 42
Quarterly Mark December 2023 - 43
Quarterly Mark December 2023 - 44
Quarterly Mark December 2023 - 45
Quarterly Mark December 2023 - 46
Quarterly Mark December 2023 - 47
Quarterly Mark December 2023 - 48
Quarterly Mark December 2023 - 49
Quarterly Mark December 2023 - 50
Quarterly Mark December 2023 - 51
Quarterly Mark December 2023 - 52
Quarterly Mark December 2023 - 53
Quarterly Mark December 2023 - 54
Quarterly Mark December 2023 - 55
Quarterly Mark December 2023 - 56
Quarterly Mark December 2023 - 57
Quarterly Mark December 2023 - 58
Quarterly Mark December 2023 - 59
Quarterly Mark December 2023 - 60
Quarterly Mark December 2023 - 61
Quarterly Mark December 2023 - 62
Quarterly Mark December 2023 - 63
Quarterly Mark December 2023 - 64
Quarterly Mark December 2023 - 65
Quarterly Mark December 2023 - 66
Quarterly Mark December 2023 - 67
Quarterly Mark December 2023 - 68
Quarterly Mark December 2023 - 69
Quarterly Mark December 2023 - 70
Quarterly Mark December 2023 - 71
Quarterly Mark December 2023 - 72
Quarterly Mark December 2023 - 73
Quarterly Mark December 2023 - 74
Quarterly Mark December 2023 - 75
Quarterly Mark December 2023 - 76
Quarterly Mark December 2023 - 77
Quarterly Mark December 2023 - 78
Quarterly Mark December 2023 - 79
Quarterly Mark December 2023 - 80
Quarterly Mark December 2023 - 81
Quarterly Mark December 2023 - 82
Quarterly Mark December 2023 - 83
Quarterly Mark December 2023 - 84
Quarterly Mark December 2023 - 85
Quarterly Mark December 2023 - 86
Quarterly Mark December 2023 - 87
Quarterly Mark December 2023 - 88
Quarterly Mark December 2023 - 89
Quarterly Mark December 2023 - 90
Quarterly Mark December 2023 - 91
Quarterly Mark December 2023 - 92
Quarterly Mark December 2023 - 93
Quarterly Mark December 2023 - 94
Quarterly Mark December 2023 - 95
Quarterly Mark December 2023 - 96
Quarterly Mark December 2023 - 97
Quarterly Mark December 2023 - 98
Quarterly Mark December 2023 - 99
Quarterly Mark December 2023 - 100
Quarterly Mark December 2023 - 101
Quarterly Mark December 2023 - Back Cover
https://www.nxtbookmedia.com