Paper: arXiv 2311.15247
Authors: HyeonJun Kim
Abstract
We investigate the information content of 3PROTV, a south Korean financial youtube channel. In our sample we found evidence for the hypothesis that the channel have information content on stock selection, but only on negative sentiment. Positively mentioned stock had pre-announcement spike followed by steep fall in stock price around announcement period. Negatively mentioned stock started underperforming around the announcement period, with underreaction dynamics in post-announcement period. In the area of market timing, we found that change of sentimental tone of 3PROTV than its historical average predicts the lead value of Korean market portfolio return. Its predictive power cannot be explained by future change in news sentiment, future short term interest rate, and future liquidity risk.
Complexity vs Empirical Score
- Math Complexity: 6.5/10
- Empirical Rigor: 6.0/10 (capped from 7.0: transaction costs not modelled; no out-of-sample test; see penalty flags)
- Quadrant: Holy Grail — high math complexity, high empirical rigor (effective empirical rigor capped from 7.0)
Why this score: The paper employs advanced econometric methods including Fama-French 5-factor models with lags/leads, event studies with CAAR calculations, and logit regressions, demonstrating substantial mathematical complexity. It also exhibits high empirical rigor through extensive data processing (YouTube scraping, sentiment lexicons), multi-year market data from DataGuide, robustness checks across parameters, and backtested trading strategies showing 11.5%-23.6% outperformance.
Research Flowchart
flowchart TD A["Research Goal:<br>Assess info content of<br>3PROTV Financial YouTube Channel"] --> B["Data Collection:<br>Video transcripts &<br>Stock price data (KOSPI)"] B --> C["Sentiment Analysis:<br>Quantify video tone<br>(Positive/Negative/Sentiment Change)"] C --> D["Analysis 1: Stock Selection<br>Event Study: Stock price<br>behavior around mentions"] C --> E["Analysis 2: Market Timing<br>Regression: Sentiment change<br>vs. Future market returns"] D --> F["Outcome A:<br>Asymmetric Price Reaction<br>Pos: Pre-announcement spike, steep fall<br>Neg: Underperformance & underreaction"] E --> G["Outcome B:<br>Predictive Power<br>Sentiment change predicts future returns<br>Independent of news, rates, & liquidity"]
Evidence Card
What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card
Proposes
The paper proposes a sentiment index derived from 3PROTV YouTube content and a trading strategy based on it.
Source sentence
We developed a sentiment index using 3PROTV’s content sentiment and found robust relation between future return and the index
Assumes 5
- Stock names mentioned more than two times are relevant to the day's script. data
Source sentence
Another attempts to lower the selection bias is to only select the stock name into our sample only when the stock name is mentioned more than 2 times.
- KOSELF lexicon accurately captures financial sentiment in Korean scripts. model
Source sentence
For sentiment analysis, we use KOSELF(Korean Sentiment Lexicon for Finance) [...]
- The scraped auto‑transcription data correctly reflects the video content. data
Source sentence
we scraped YouTube’s auto-transcription data to extract the mentioned stock and script’s overall sentiment through August 2, 2022 to November 28, 2022, which in total a 500 content scripts.
- The sentiment index S(N,t) can be binary‑coded as defined in equation (6). model
Source sentence
We define a sentiment index S(N, t) with a binary value that is defined as equation 6.
Show 1 more
- Future news sentiment, short‑term interest rates and liquidity risk do not fully explain the predictive power of the sentiment index. statistical
Source sentence
Its predictive power cannot be explained by future change in news sentiment, future short term interest rate, and future liquidity risk.
Data
- 3PROTV YouTube auto‑transcription scripts text-news public
Source sentence
we scraped YouTube’s auto-transcription data to extract the mentioned stock and script’s overall sentiment through August 2, 2022 to November 28, 2022, which in total a 500 content scripts.
- DataGuide Korean firm name list alternative unknown
Source sentence
We then match the tokenized words of the script with DataGuide’s Korean firm name list.
- KOSELF Korean Sentiment Lexicon for Finance alternative unknown
Source sentence
For sentiment analysis, we use KOSELF(Korean Sentiment Lexicon for Finance) [...]
Costs
Not modeled
Source sentence
The daily trading strategy using our sentiment index show outperformance from 11.5% to 23.6% without transaction cost.
Finds 5
- Negatively mentioned stocks underperform after the announcement period. performance
Source sentence
Negatively mentioned stock started underperforming around the announcement period, with underreaction dynamics in post-announcement period.
- Positively mentioned stocks exhibit a pre‑announcement price spike followed by a steep fall around the announcement. performance
Source sentence
Positively mentioned stock had pre-announcement spike followed by steep fall in stock price around announcement period.
- The sentiment index S(N,t) predicts the sign of 2‑day leading market portfolio return. predictive-accuracy
Source sentence
we develop a preprocessed sentiment index S(N, t) from SENTIMENT (i), which we demonstrate that it has predictive power for sign of 2day leading market portfolio return.
- A trading strategy based on the sentiment index yields 11.5%–23.6% outperformance without transaction costs. performance
Source sentence
The daily trading strategy using our sentiment index show outperformance from 11.5% to 23.6% without transaction cost.
- Predictive power of the sentiment index remains after controlling for future news sentiment, short‑term interest rates, and liquidity risk. statistical
Source sentence
The predictability persisted even controlling for future news sentiment, short term interest rate, and liquidity risk factor (PctZero), making 3PROTV’s content information somewhat independent from those information.
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 26 related papers). Coverage is partial, so this is neither replication nor refutation.
Code
No repository by the authors found in the paper.
Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.
Cited References (12)
- The information content of annual earnings announcements (1968)
- The information content of the president’s letter to shareholders (1996)
- Information content of equity analyst reports (2005)
- Trading on twitter: The financial information content of emotion in social media (2014)
- Trading on twitter: Using social media sentiment to predict stock returns (2017)
- Event studies: A review of issues and methodology (1989)
- A five-factor asset pricing model (2015)
- Building the Korean sentiment lexicon for finance (KOSELF) (2021)
and 4 more
- Investor sentiment in the stock market (2007)
- Machine-Learning-Based News Sentiment Index (NSI) of Korea (2022)
- The information content of the sentiment index (2016)
- The disposition effect and momentum (2002)
0 of 12 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.