Principal researchers: Dr. Lucia Alessi (JRC) and Dr. Mikhail Oet (Northeastern University)

JRC-NEU XN Project 1: Multi-level

sentiment analysis of stress

The project fills the unmet need of companies and individuals to measure unobservable financial stress in networks of multiple levels, including economic sectors and types of financial instruments. textual information, such as sentiment expressed about shifts in the economy and adverse The project develops a multi-level measurement of financial stress based on conditions in various types of financial intermediaries.

Abstract

Problem: At the firm level, financial stress can produce layoffs and bankruptcy. At an instrumental level, we may face a stock market crash. At the state level, there are consequences like funding cuts to education and government programs. Itcreate a theory to measure the multi-level financial stress. is necessary to

Motivation: Understanding financial stress can help people avoid dire economic consequences. Unexpected financial pressures can lead individuals and companies to hoard cash and liquidate assets. By understanding endogenous factors of financial stress, we will have better tools to prevent such distress from threatening the national economy. Methods: capture the sentiment variance across the national population regulatory and investing Our sentiment analysis approach synthesizes multi-level textual publications to community and the general public. News Articles undergo a Lexicon-based sentiment analysis and machine learning algorithm of supervised classification. These measures are validated empirically by explaining observed actions across firm and instrument types, controlling for macroeconomic conditions, changing preferences for liquidity, and the time horizon. Results/Conclusion: Our research introduces two important contributions to knowledge: 1) we propose and validate that expressed sentiment propagates uncertain financial stress formation across multiple levelsexpressed sentiment explains the volatility in the response to financial stressors across multiple of financial instruments and sectors, 2) we propose and validate that levels.

Principal researchers: Dr. Lucia Alessi (JRC) and Dr. Mikhail Oet (Northeastern University)

1. Introduction

1.1. Introduction to the Problem of Practice

Financial stress is symptom of economic conditions. As conditions change, economic euphoria may
be followed by a shock. Healthy assets become toxic. Healthy firms can become weak. An economy
weakened by prolonged financial stress can deteriorate to a state of severe recession. Due to the
potential risks that prolonged exposure to financial stress sets to the economy, there is a need for
individuals and companies to measure the unobserved financial stress in networks of multiple levels,
including economic sectors and types of financial instruments. Financial stress occurs when the
commercial systems strain and their ability to arbitrate is impaired (Hakkio & Keeton, 2009).
According to Cevik et al. (2012), this condition occurs due to the number of factors such as significant
shifts in prices of assets, the uncertainty of the future value of assets, reduced tendency to hold high-
risk assets and liquidity droughts with a few willing to hold illiquid assets. The major players in the
financial system are the: Markets, which include equity, bond, money market and foreign exchange
markets; Intermediaries: banks, insurance and Infrastructure comprised of payments, settlements and
clearing systems. (Krishnamoorthy, 2018)
This paper is structured as follows .We first highlight our research motivations by explaining the
problem of practice and why it is relevant, then propose our research questions. Our thesis is
explicitly stated and our data sources, evidence, research methods, and expected findings follow.
Lastly, we provide an overview of potential rebuttals and then conclude the proposal.

1.2. Research Motivation and Research Backing

It is undisputed that negative economic outcomes are characterized by financial stress. The problem
of financial stress can affect our personal and professional lives directly and immediately. If financial
stress is placed on a firm to a point where layoffs are required, you or your favorite coworker could be
left without a job.
When economic times are bad, you can typically tell by the news reported in the moment of instability

• the same goes for prosperous economies. However, what might not be so obvious is how the news

reflects the financial stress that is faced by an economy, and how it causes individuals and institutions to take action. At an aggregate level of an entire economy, we can estimate financial stress in a variety of ways through the use of some useful and representative indicators. However, because financial stress does not have a universal definition, it is currently not possible to measure financial stress in detail for the microeconomic atoms of the economy, such as individual economic agents or specific financial

Principal researchers: Dr. Lucia Alessi (JRC) and Dr. Mikhail Oet (Northeastern University)

instruments in use. We define the concept of financial stress simply as the difference between financial stability and financial instability, two other related concepts. Therefore, we simplify the problem of measurement of the construct of financial stress in terms of the measurement of the constructs of financial stability and financial instability. To operationalize this microeconomic research on financial stress, we estimate the stability and instability constructs in specific types of financial agents as they are discussed by the textual publications used by governments, firms, and individuals to make decisions.

The European Central Bank highlights the need for additional tools to counteract financial instability (stress) in the paper, “Macroprudential policy at the ECB: Institutional framework, strategy, analytical tools and policies” (Constâncio, V., 2020). We propose a component of a new tool for ECB, one that derives sentiment’s effect on financial instability.

This paper aims to discuss whether policymakers' discussion of financial stability and other factors can systematically explain the observed deviation between the policy rate and the interest rate implied by the Taylor rule. According to Krippendorff (1989: 403), “content analysis is a research technique for making replicable and valid inferences from data to their context.” Stemler (2001) defines it as a systematic methodology “for compressing text into content categories based on explicit rules of coding.” Since the text of the FOMC meetings is one of our data sources, the content analysis should also be used.

Based on Oet, M.V.,et al. (2013), “the risk data is based both on publicly available financial information and on private supervisory EWS of individua institutions’ risk. The economic- value time series is obtained through private supervisory FRB-IRR Focus and FRB-Bank CaR (Frye and Pelz, 2008 ) models”. In addition, the paper discusses some financial contagion through various channels of transmission. For example, direct contagion through interbank credit and liquidity markets and indirect contagion resulting from a general deterioration in financial market conditions.

With the diversity of studies in the domain of Sentimental analysis, this study focuses on the financial sector. Our method of analysis focuses on texts which can be interpreted by Machine Learning (ML) methods by making use of documents used to “learn” and specific financial features to classify the words. Some of the common features used in engineering schemes rely on internal text features such as bag-of-words, Part-of-Speech (POS) tags and Named Entities (Antweiler & Frank, 2005; Schumaker et al., 2012). Other studies add polarities from lexicons as features of vectors (Bollen et al., 2011) and assign sentiment classes at the sentence level. In the process of vocabulary classification, the problem of class-imbalance may arise. Class - imbalance is refers to the classification task in different categories of training sample number difference is very big. We may be able to solve this problem by expanding the data set and trying different classification algorithms. More data often trumps better algorithms. Because machine learning uses existing data to estimate the distribution of the entire data, more data can often yield more distribution information. Even if the addition of small-category sample data increases the addition of large-category sample data, it can be solved by giving up part of the large-category data (i.e. undersampling of the large-category data).

Principal researchers: Dr. Lucia Alessi (JRC) and Dr. Mikhail Oet (Northeastern University)

Because different algorithms are suitable for different tasks and data, different algorithms should be used for comparison. Decision trees tend to do well on categorically unbalanced data. It uses class variability-based partitioning rules to create a classification tree so that samples from different categories can be forcibly separated.

Among related literature using similar methods and literature, our study applies and extends the approaches of Smith (date) and Garcia (2013). Smith’s “Feeding Unrest: Disentangling the causal relationship between food price shocks and sociopolitical conflict in urban Africa” uses the GDELT database for accounting for events. This paper is a good reference for us for the use of national instrumental variables and distinguishing causal relationships. Their theory, specifically, looks at the causality between civil unrest and food prices. Garcia’s “Sentiment during Recessions” uses New York Times financial articles between 1905 and 2005 to study the effects of sentiment on asset prices (2013). This study uses positive and negative word counts to assign sentiment scores - ideally, we will be using phrases for sentiment scores. One estimate in this study is estimating the effect that sentiment has on the volume of abnormal stock trades (p. 1294, Garcia, 2013).

1.3. Research Questions

Given our motivation and backing, we have several research questions to consider:

● Can otherwise undetectable financial stress be measured through sentiment expressed in
publications?
● Does negative sentiment from financial news articles correlate with financial instability?
● To what extent does negative sentiment from financial news perpetuate a negative effect on
financial markets?
● Can we determine the lag of the effects of sentiment from FOMC meeting minutes, online
articles, and article comments on financial markets?
● Does the sentiment about certain market segments have varying weights on financial stress
than does the sentiment from another market segment?
● Are our findings transferable to more than just financial markets?

1.4. Hypotheses

H1) associated with financial stress expressed through sentiment of articles related to United States Present and future value of assets, uncertainty on stock, bond, and forex markets are Chartered Depository Institutions.H2) The asymmetry of information in the interbank system and stock exchange is associated with financial stress.H3) Decreased willingness to hold risky and illiquid assets are associated with financial stress.

Principal researchers: Dr. Lucia Alessi (JRC) and Dr. Mikhail Oet (Northeastern University)

H4) the forex rate are associated with financial stress. Economic conditions such as GDP growth, price stability, sustainable debt, stability in

This research project, therefore, intends to develop a multi-level measurement of financial stress based on textual information, such as sentiments about shifts in the economy and adverse conditions in various types of financial intermediaries.

Sentimental analysis is concerned with judgments, response and emotions generated from texts, widely used in fields such as data mining, web mining and social media analytics. Sentiment analysis system for textual analytics merges natural language processing (NLP) and machine learning techniques in assigning weighted-sentiments to score to an entity, topic, theme and a category within a sentence or a phrase. Sentiments made are usually used to judge one’s perspective and one's behavior and can either be positive, negative or neutral or can contain an arithmetic score of a scale of say 1 to 5. Sentimental analysis discovers opinions, classifies the attitude conveyed and categorizes into a gives scores on the level. Classification may be done by machine learning technique, lexicon-based which uses the polarity score of a given text or a combination of both. Finally, evaluation metrics of precision, Recall, F-score and or Accuracy is obtained to evaluate how the accuracy of the analysis.

1.5. Claim and Qualifiers

Financial stress can be represented and predicted, to an extent, by the sentiment of text that is published at the individual, firm, and governmental level. We know there is a connection between stress and human response. “The cognitive activation theory of stress (CATS)” theorizes that our body has an alarm system in response to stressful stimuli (Ursin, 2003). Depending on the nature of the stimuli, there is either a positive or negative response. We expect that stimuli of positive and negative sentiment both reflect the states of economies and can cause consumers to respond.

The essence of our research methods stems from papers like “Detecting Risks in the Banking System by Sentiment Analysis” (Nopp, 2015), “Big Data and Economic Forecasting: A Top-Down Approach Using Directed Algorithmic Text Analysis” (Nyman, 2018), and “News and Narratives in Financial Systems: Exploiting Big Data for Systemic Risk Assessment” (Nyman, n.d.). A common method used in these papers is the directed algorithmic text analysis (DATA).

In Nopp’s paper, there is an implied claim that banking supervisors can use sentiment analysis to derive valuable information regarding risk assessments. This research used 500+ CEO Letters and Outlook Sections from bank annual reports to provide evidence for their claim. In Nyman’s two papers, the DATA method is used to forecast economic events.

Principal researchers: Dr. Lucia Alessi (JRC) and Dr. Mikhail Oet (Northeastern University)

Nyman and colleagues importantly use Conviction Narrative Theory (CNV) as a backing to their
claim. CNV is the proposal that a combination of narrative and emotion combine to facilitate action by
individuals (Tuckett, 2013). These choices can be linked to economic outcomes. According to
Tuckett, conviction narratives contain two elements: attractiveness and “something to manage doubt
[of the narrative].”
The difference between our paper and Nyman, Nopp, and Ursin’s, is that we identify a latent variable
as a connection between stress received through literature and economic outcomes (financial stress).

1.6. Data Sources

We retrieve news articles from the Global Database of Events, Language, and Tone (GDELT) Global
Knowledge Graph (GKG)1.0 (GDELT, 2020). GDELT GKG 1.0 provides us with web news in textual
format. Data from the prior day is made available at 6am EST of the current day. Data with our search
criteria dates back to 2/20/2015. This indexed text covers online news which GDELT has analyzed for
mentioned themes and organizations as well as tone, or sentiment, of the article (GDELT, 2020).

1.7. Research Methods

Our exploratory analysis is handled during text collection and sentiment analysis while explanatory analysis occurs through MIMIC structural modeling.

We first collect our URLs and text for sentiment analysis from GDELT. Our GDELT articles are constrained from the date parameter of 2/20/2015 to the present time. We constrained results to the United States solely and we built a representative list with the top 40 US commercial banking organizations and the top 40 insurance organizations. Organization names were formatted for a python search as shown in Figure 1.71. See Appendix 1 — Organizations for full organizations list. Our formatted search criteria are available upon request.

We retrieve our GDELT GKG data using python. The GKG 1.0 database allowed us to filter results by location, themes, and organizations, with the ultimate goal of obtaining an article SOURCEURL.

GDELT Date^1 : [20150220:20200303]

GDELT Location^2 : #US

1 2 https://www.gdeltproject.org/data.html#documentation^ https://www.gdeltproject.org/data.html#documentation

Principal researchers: Dr. Lucia Alessi (JRC) and Dr. Mikhail Oet (Northeastern University)

GDELT Themes^3 : ['WB_1234_BANKING_INSTITUTIONS','WB_1236_COMMERCIAL_BANKING','WB_1256_CREDIT_U NIONS','ECON_DEBT','ECON_STOCK_MARKET','WB_1234_BANKING_INSTITUTIONS','WB_1920_ FINANCIAL_SECTOR_DEVELOPMENT','ECON_CENTRALBANK','WB_318_FINANCIAL_ARCHITEC TURE_AND_BANKING','WB_332_CAPITAL_MARKETS','WB_611_PENSION_FUNDS','WB_971_BAN KING_REGULATION']

GDELT Organizations^4 : See 4.1 Appendix

Figure 1.71 Organizations Sample

Organization Ticker Searchable Format
Equity Bancshares Inc EQBK ‘equity bancshares’
Esquire Financial Holdings Inc ESQ ‘esquire financial holdings’
Evans Bancorp, Inc. EVBN ‘evans bancorp’
F & M Bank Corp. FMBM ‘f m bank’
F.N.B. Corporation FNB ‘fnb corporation’

Each sector is notated by the following Sector ID. There were six sectors with overlapping organization names which were combined into a single GDELT query, thus the results of those are aggregated.

Figure 1.72 Sector Descriptions

Sector ID Sector Description
11 US Chartered Depository Institutions

3 4 https://www.gdeltproject.org/data.html#documentation^ https://www.gdeltproject.org/data.html#documentation

Principal researchers: Dr. Lucia Alessi (JRC) and Dr. Mikhail Oet (Northeastern University)

14 Credit Unions
15 Property Casualty Insurance Companies*
16 Life Insurance Companies*
21 Money Market Funds**
22 Mutual Funds**
23 Closed End Funds
24 Exchange Traded Funds
25 Government Sponsored Enterprises***
26 Agency and GSE Backed Mortgage Pools***
27 Issuers of Asset Backed Securities
29 Real Estate Investment Trusts (REITs)
30 Security Brokers and Dealers
31 Holding Companies

The results of the GDELT query retrieves 1,682,785 articles. For each sector, we randomized the aggregated articles and scraped a sample of 6,000 articles for each sector aside from REITs.

Figure 1.73 Sample Outline

Sector ID Sector Description

GDELT Total
Article Count SampleScrape

Count of Scraped
Articles ≤ 75 Total
Words

% Error Total
Scraped

Indexed Articles
Samplefrom ArticlesGDELT % of
11 US Chartered Depository Institutions 320,440 6,000 912 0.152 5,088 0.
14 Credit Unions 38,936 6,000 1,039 0.173 4,961 0.
15_16 Property Casualty Insurance Companies* +Life Insurance Companies* 42,337 6,000 929 0.155 5,071 0.
21_22 Money Market Funds +Mutual Funds** 353,567 6,000 1,101 0.184 4,899 0.
23 Closed End Funds 38,595 6,000 1,133 0.189 4,867 0.
24 Exchange Traded Funds 22,450 6,000 1,214 0.202 4,786 0.

Principal researchers: Dr. Lucia Alessi (JRC) and Dr. Mikhail Oet (Northeastern University)

25_26 Government Sponsored Enterprises +Agency and GSE Backed Mortgage Pools***^ 32,401 6,000 1,069 0.178 4,931 0.
27 Issuers of Asset Backed Securities 260,061 6,000 1012 0.169 4,988 0.
29 Real Estate Investment Trusts (REITs) 12,837 2,869 343 0.12 2,526 0.
30 Security Brokers and Dealers 164,393 6,000 1,129 0.188 4,871 0.
31 Holding Companies 396,768 6,000 1,042 0.174 4,958 0.
Totals - 1,682,785 62,869 10,923 0.174 51,946 0.

The remaining articles with a scraped word count of greater than or equal to 75 are denoted indexed data. We verified that this was a good representation of the GDELT population as well as the original sample with a frequency histogram.

GDELT Article Counts by Sector exhibits the fluctuations in total articles for each of our sector searches over time.

Principal researchers: Dr. Lucia Alessi (JRC) and Dr. Mikhail Oet (Northeastern University)

Principal researchers: Dr. Lucia Alessi (JRC) and Dr. Mikhail Oet (Northeastern University)

In order to measure sentiment, we build a sentiment index because we need a quantitative measure of the qualitative construct. We perform lexicon-based sentiment tagging and develop our index as an RSS, Relative Sentiment Shift, which was employed in “News and Narratives in Financial Systems: Exploiting Big Data for Systemic Risk Assessment.” (Nyman, 2018). Each individual article was analyzed through python. Our sentiment index is constructed below.

Sentiment = (Excitement - Anxiety) / Total Words

We pursue the sentiment analysis methods like DATA and those used for “Detecting Risks in the Banking System by Sentiment Analysis” (Nopp, 2015). These methods include Lexicon-based sentiment analysis (sentiment tagging, term weighting, valence shifting, calculating sentiment scores) and machine learning algorithm of supervised classification (Assigning the Class Labels, Preprocessing, Feature Selection, and Classification). The collected, generated, and merged data is classified as positive or negative according to the classification rules. The results allow empirical testing of our hypotheses.

Principal researchers: Dr. Lucia Alessi (JRC) and Dr. Mikhail Oet (Northeastern University)

Principal researchers: Dr. Lucia Alessi (JRC) and Dr. Mikhail Oet (Northeastern University)

Principal researchers: Dr. Lucia Alessi (JRC) and Dr. Mikhail Oet (Northeastern University)

We build our outcome variables using the Financial Accounts of the United States from the Board of Governors of the Federal Reserve.

Results pending

Multiple Indicator–Multiple Cause (MIMIC) Models are applied to our data as there are several variables at multiple levels affecting one another. Both the MIMIC model and the Multiple Indicator models generate a correlation matrix for us to use to see the correlation coefficients. We will create a causal graph like those seen in the MIMIC model.

MIMIC model is a special kind of structural equation model. Its explanatory variables are a set of explicit variables (causes), while the explained variables are hidden variables (defined or measured by several indicator variables). The use of a set of explanatory variables to predict one or more potential variables. A MIMIC model can have several interpreted variables, not just one.

Principal researchers: Dr. Lucia Alessi (JRC) and Dr. Mikhail Oet (Northeastern University)

Figure 1: Sample MIMIC Model

Results pending

We use the Granger causality approach to test the relationship between variables. In time series, the two economic variables’ Granger causality relation between X and Y is defined as: if the contains information about the variables X and Y past condition, the effect is better than for the forecast of variable Y only separate by Y past information to the prediction effect of Y, which variables X helps to explain the change in the future, Y is considered variable X is the Granger reason of variable Y. Although Granger causality does not equal actual causality, it is statistically significant for economic forecasting.

According to Krippendorff (1989: 403), “content analysis is a research technique for making replicable and valid inferences from data to their context.” Stemler (2001) defines it as a systematic methodology “for compressing text into content categories based on explicit rules of coding.”

1.8. Methodological Challenges

The MIMIC model aids in detecting any methodological challenges of identification and confounding constructs. We need to know that we are measuring the parameters that we intend to measure. To avoid an identification problem, we must find the best estimate of our parameter values.

Confounding constructs occur when there is a variable influencing both the dependent and independent variables. Construct validity is to be considered in our methodological challenges. To gain construct validity, we should show that our measures of financial stress and economic outcomes are interrelated, but try to narrow our scope enough to eliminate confounding variables. Alternative causal variables must be considered in order to better establish causality between sentiment and financial stress.

Principal researchers: Dr. Lucia Alessi (JRC) and Dr. Mikhail Oet (Northeastern University)

Instrumental variables are introduced to the model when we can measure the effect one variable is having on another. A problem with these variables is that we will need a formal instrument, like cash flow, in order to introduce these variables into our models.

1.9. Expected Findings

Once our theory is supported, we discuss the extent in which our findings are transferable to other
disciplines.
In our research we find that sentiment, measured as an index, from official central bank reports,
analyst reports, news articles, and world events can measure financial stress.
We find a correlation between negative sentiment and financial instability. Sentiment from central
bank reports and analyst reports lead sentiment from private news and world events. The sentiment
from bank and analyst reports will lead overall financial stress.
There is be a cyclical effect in our sentiment findings: the greater the negative sentiment in news, the
greater portion of news with negative sentiment will follow. We will have to account for this
phenomenon.
Last, we find that sentiment in some market segments has greater variability than do others, and
sentiment in some market segments has a greater effect on the market than do others.

1.10. Counter-Arguments

We should note that bias is definitely possible in our methodology because language in general can
hold biases. For example, the sentiment analysis algorithm is supervised, it may be biased depending
on the supervisor.
Another potential counter-argument to our findings is symbolic interactionism. Many choices that are
made are significant to the individual who makes the choice and stress-related choices are no
different. When individuals or firms are making a choice that’s going to affect their stress level, they
may attach a personal importance to their decision and this may cause them to behave economically
irrationally. For example, a company CEO or company board may not want to disappoint their
shareholders, so they release a positive article about their financial standing. The leadership of this
company attaches more importance to the shareholders rather than the net benefit to the economy.
A final potential weakness that can cause bias in data is the existing principal agent problem between
news outlets and news consumers. Consumers want knowledge that will help them make

Principal researchers: Dr. Lucia Alessi (JRC) and Dr. Mikhail Oet (Northeastern University)

economically beneficial decisions. News organizations have incentives to produce news that is more attractive to consumers, without needing to consider if it’s in consumers’ best interest.

1.11. Conclusion

In conclusion, we have laid out a plan to detect and identify the latent variable of financial stress through sentiment analysis and the causal model. Our research motivations provide an important reasoning for our efforts: financial stress causes negative economic outcomes. Our research questions give us interesting questions to explore during our research, that will limit the scope of our findings. We highlighted our key data sources, research methods, and expected findings. Lastly, we will have provided limitations that can weaken our theory.

1.12. References

Antweiler, W., & Frank, M. Z. (2005, November 27). Is All That Talk Just Noise? The Information Content of Internet Stock Message Boards. Retrieved from https://onlinelibrary.wiley.com/doi/10.1111/j.1540-6261.2004.00662.x

Federal Open Market Committee. (2020). Meeting calendars, statements, and minutes (2015-Federal Reserve. Retrieved from 2020). https://www.federalreserve.gov/monetarypolicy/fomccalendars.htm.

European Central Bank. (2020). ECB Research & Publications. Retrieved from https://www.ecb.europa.eu/pub/html/index.en.html

Constâncio, V., Cabral, I. Detken, C., et. al. (July 2019) Macroprudential policy at the ECB: Institutional framework, strategy, analytical tools and policies.

Global Database of Events, Language, and Tone (GDELT). (2020). The GDELT Project. Retfrom https://www.gdeltproject.org/data.html#googlebigquery rieved

GARCÍA, D. (2013). Sentiment during Recessions. The Journal of Finance, 68(3), 1267-Retrieved from http://www.jstor.org/stable/42002620 1300.

Krishnamoorthy, S. (2018). Sentiment analysis of financial news articles using performance indicators. Knowledge and Information Systems , 56 (2), 373-394. (Secondary data source. https://www.groundai.com/project/sentiment-analysis-of-financial-news-articles-using-

Principal researchers: Dr. Lucia Alessi (JRC) and Dr. Mikhail Oet (Northeastern University)

performancGARCÍA, D. (2013). Sentiment during Recessions. The Journal of Finance, 68(3), 1267-1300. Retrieved from http://www.jstor.org/stable/42002620e-indicators/1).

Multiple Indicator–Multiple Indicator Cause, Mixture, and Multilevel Models. (n.d.).
Nopp, C., & Hanbury, A. (2015). Detecting Risks in the Banking System by Sentiment Analysis. Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing,
591–600. doi: 10.18653/v1/d15- 1071
Nyman, R. M., Kapadia, S. M., Tuckett, D. M., Gregory, D. M., Ormerod, P. M., & Smith, R. M. (2018). News and Narratives in Financial Systems: Exploiting Big Data for Systemic Risk Assessment.
SSRN Electronic Journal. doi: 10.2139/ssrn.
Nyman, R., Ormerod, P., Smith, R., & Tuckett, D. (n.d.). Big Data and Economic Forecasting: A Top-Down Approach Using Directed Algorithmic Text Analysis. Centre for the Study of Decision-
Making Uncertainty.
Stemler, S. 2001. An overview of content analysis. Practical Assessment, Research and Evaluati7: 137-146. on,

Smith, T. (2014). Feeding unrest: Disentangling the causal relationship between food price shocks and sociopolitical conflict in urban Africa. Journal of Peace Research, 51(6), 679-695.
Retrieved from http://www.jstor.org/stable/
Tsigos, Cand Pathophysiology.., Kyrou, I., Kassi, E., & Chrousos, G. P. (2016, March 10). Stress, Endocrine Physiology

Tuckett, D. (April, 2013). Irreducible Uncertainty and its Implications: A Narrative Action Theory for Economics. INET Hong Kong April 2013. Retrieved from
https://www.ineteconomics.org/uploads/papers/Tuckett-Paper.pdf
Ursin, H., & Eriksen, H. R. (2004). The cognitive activation theory of stress. Psychoneuroendocrinology , 29 (5), 567–592. doi: 10.1016/s0306-4530(03)00091-x

V., M., Lyytinen, & Kalle. (2017, January 30). Does Financial Stability Matter to the Fed in Setting US Monetary Policy? *. Retrieved from https://academic.oup.com/rof/article-
abstract/21/1/389/296453 1

2. Literature Review

Principal researchers: Dr. Lucia Alessi (JRC) and Dr. Mikhail Oet (Northeastern University)

2.1. Claims and Hypotheses Literature^5

Past studies have focused on analyzing texts and news from digital sources and

then explore the effects on the economy. An example is Romer & Romer (2010) who

analyzed text sources from speeches to assess the probable effects of changes in

taxes on economic status. Dominguez & Shapiro (2013) analyzed newspaper and

media information to establish events propelling shifts in the Euro crisis and how it

accounted for the slowdown in the economic recovery. Soo (2013) measured the

positive and negative tone of housing news from newspaper articles to test the

capability of sentiments in the run-up and crash of housing prices that inspired the

US financial crisis of 2008.

2.2. Methods Literature^6

With the diversity of studies in the domain of Sentimental analysis, this study

focuses on the financial sector. Our method of analysis focuses on texts which can

be interpreted by Machine Learning (ML) methods by making use of documents

used to “learn” and specific financial features to classify the words. Some of the

common features used in engineering schemes rely on internal text features such as

bag-of-words, Part-of-Speech (POS) tags and Named Entities (Antweiler & Frank,

2005; Schumaker et al., 2012). Other studies add polarities from lexicons as

features of vectors (Bollen et al., 2011) and assign sentiment classes at the

sentence level.

2.3. Data Literature^7

2.3.1. Country Level Data:

• https://www.federalreserve.gov/monetarypolicy/fomccalendars.htm
• https://www.gdeltproject.org/

Smith’s “Feeding Unrest: Disentangling the causal relationship between food price

shocks and sociopolitical conflict in urban Africa” uses the GDELT database for

accounting for events. This paper is a good reference for us for the use of national

(^5) 1. Your claims and hypotheses for the testable propositions. Have these propositions already been^ tested? What were the conclusions? What are the expected signs for (^6) Your proposed methods. Have your methods been used for a similar problem? What limitations arise? the relationships? In summary, how were they handled technically? 7 Regarding the data you want to use, has this data been used for similar research by others?

Principal researchers: Dr. Lucia Alessi (JRC) and Dr. Mikhail Oet (Northeastern University)

instrumental variables and distinguishing causal relationships. Their theory,

specifically, looks at the causality between civil unrest and food prices.

2.3.2. Firm Level Data:

Garcia’s “Sentiment during Recessions” uses New York Times financial articles

between 1905 and 2005 to study the effects of sentiment on asset prices (2013).

This study uses positive and negative word counts to assign sentiment scores -

ideally, we will be using phrases for sentiment scores. One estimate in this study is

estimating the effect that sentiment has on the volume of abnormal stock trades (p.

1294, Garcia, 2013).

2.3.3. Individual Level Data:

• https://www.gdeltproject.org/

Many studies have used twitter data for sentiment analysis. There is an early use of

sentiment analysis from twitter in “Cognition, market sentiment and financial

instability” (Dow, 2008). They control their sentiment model using events occurring

around the recession and 2008 election.

2.4. Novel Approaches^8

We have not determined our approaches to be novel.

2.5. References

Antweiler, W., & Frank, M. Z. (2005, November 27). Is All That Talk Just Noise? The

Information Content of Internet Stock Message Boards. Retrieved from

https://onlinelibrary.wiley.com/doi/10.1111/j.1540-6261.2004.00662.x

Bollen, J., Mao, H., & Pepe, A. (2011).. In International AAAI Conference on Web

and Social Media. Retrieved from

https://www.aaai.org/ocs/index.php/ICWSM/ICWSM11/paper/view/2826/

8

for different problems? 4. If you are proposing novel approaches, have similar approaches been used for related problems or

Principal researchers: Dr. Lucia Alessi (JRC) and Dr. Mikhail Oet (Northeastern University)

Dominguez, M.E., K., Shapiro, & D., M. (2013, February 2). Forecasting the

Recovery from the Great Recession: is this Time Different? Retrieved from

https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2210771

Dow, S. (2011). Cognition, market sentiment and financial instability, Cambridge

Journal of Economics , Volume 35, Issue 2, March 2011, Pages 233–249.

Retrieved from https://doi.org/10.1093/cje/beq029

Garcia, D. (2013). Sentiment during Recessions. The Journal of Finance, 68(3),

1267-1300. Retrieved from http://www.jstor.org/stable/42002620

Smith, T. (2014). Feeding unrest: Disentangling the causal relationship between

food price shocks and sociopolitical conflict in urban Africa. Journal of Peace

Research, 51 (6), 679-695. Retrieved from http://www.jstor.org/stable/24557493

Romer, Christina, Romer, & H., D. (n.d.). The Macroeconomic Effects of Tax

Changes: Estimates Based on a New Measure of Fiscal Shocks. Retrieved

from https://www.aeaweb.org/articles?id=10.1257/aer.100.3.763

Oet, M. V., Bianco, T., Gramlich, D., & Ong, S. J. (2013). SAFE: An early warning

system for systemic banking risk. Journal of Banking & Finance, 37(11), 4510-

4533 Retrieved from

https://www.sciencedirect.com/science/article/pii/S0378426613001040

Oet, M., Lyytinen, K. (2017). Does Financial Stability Matter to the Fed in Setting US

Monetary Policy?, Review of Finance , Volume 21, Issue 1, March 2017, Pages

389 –432. Retrieved from https://doi.org/10.1093/rof/rfw054

Schumaker, R. P., Zhang, Y., Huang, C.-N., & Chen, H. (2012). Evaluating

sentiment in financial news articles. Decision Support Systems, 53(3), 458–

464. doi: 10.1016/j.dss.2012.03.001 Retrieved from

http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.457.6544&rep=rep1&

type=pdf

Seth, A. (n.d.). Granger causality. Retrieved from

http://www.scholarpedia.org/article/Granger_causality

Soo, C. (2013). ‘Quantifying Animal Spirits: News Media and Sentiment in the

Housing Market. What type of publication and from where?’, Ross School of

Principal researchers: Dr. Lucia Alessi (JRC) and Dr. Mikhail Oet (Northeastern University)

Business Paper no. 1200 Retrieved from

https://www.researchgate.net/publication/272290660_Quantifying_Animal_Spiri

ts_News_Media_and_Sentiment_in_the_Housing_Market

Wang, X., & Wang, J. (n.d.). Structural Equation Modeling: Applications Using Mplus.

Retrieved from https://www.oreilly.com/library/view/structural-equation-

modeling/9781118356302/c03anchor-1.html#:~:text=MIMIC model stands for

multiple,) affect latent variables/factors.

3. Data Selection

3.1. Tutorial Information

GDELT Example & Tutorial

https://github.com/ttakko/newsdataexamples/blob/master/gdelt_example.ipynb

1. https://pypi.org/project/gdelt/
2. https://nbviewer.jupyter.org/github/JamesPHoughton/Published_Blog_Scripts/blob/master/GDELT%20Wrangler%20-%20Clean.ipynb
3. https://github.com/linwoodc3/gdeltPyR
4. https://towardsdatascience.com/making-f171d1da65a8 sense-of-the-news-part-1-introduction-

How to access an API with Key/Passcode.

https://www.google.com/search?q=how+to+use+an+api+key&rlz=1C5CHFA_enUS789US789&oq=how+t o+use+an+API+key&aqs=chrome.0.0l8.3427j0j7&sourceid=chrome&ie=UTF- 8#kpvalbx=_QL9JXpG2NY2rytMPhJWToA838

3.2. Country Level Data

GDELT

Principal researchers: Dr. Lucia Alessi (JRC) and Dr. Mikhail Oet (Northeastern University)

Macroeconomic events:

https://www.gdeltproject.org/data.html#gdeltanalysisservice

The GDELT Analysis service:

https://www.gdeltproject.org/data.html#googlebigquery

Federal Reserve Federal Open Market Committee

Current Meeting Minutes:

https://www.federalreserve.gov/monetarypolicy/fomccalendars.htm

Historical Meeting Minutes:

https://www.federalreserve.gov/monetarypolicy/fomc_historical_year.htm

Minutes: available back to 1936. They are a summary of the addressed policy issues by meeting participants. We The minutes of each regularly scheduled meeting of the Federal Open Market Committee are will need to download these minutes individually.

3.3. Instrumental Level Data

Financial Accounts of the United States

https://www.federalreserve.gov/releases/z1/20191212/html/d1.htm

Debt Growth By Sector -

● Domestic financial sectors
● Domestic nonfinancial sectors

Principal researchers: Dr. Lucia Alessi (JRC) and Dr. Mikhail Oet (Northeastern University)

3.4. Firm Level Data

New York Times

https://developer.nytimes.com/

https://developer.nytimes.com/docs/archive-product/1/overview

*Created developer account in New York Times, Public API’s all available for use

App Name: CED 6090

Description: multi-level sentiment analysis

App ID: f9412b30-dc06-41f3-b54e-74f75e473642

Key: **********************************

Features:

Article Search: Look up articles by keyword. You can refine your search using filters and facets.

Semantic API: Use to search “Articles” by keywords and descriptors.

Nexis Uni

https://library.northeastern.edu/research/resources/items/nexis-uni

Nexis Uni features more than 15,000 news, business and legal sources from LexisNexis available through northeastern. This site will require sampling with the 100 document limit on downloading files. Aside from key word search, the parameters are:

● Timeline
● Publication Type
● Subject

Principal researchers: Dr. Lucia Alessi (JRC) and Dr. Mikhail Oet (Northeastern University)

● Industry
● Geography by Document
● Negative News
● Sources
● Language
● People

EX: when I search the term “credit market” there are more than 10,000 results from Nexis Uni. Nexis gives us a frequency histogram for the Timeline parameter.

3.5. Individual Level Data

New York Times

https://developer.nytimes.com/docs/archive-product/1/overview

Community API: comments from registered users on New York Times articles. NOTE: The Community API is beta.

3.1.1.a) What data is used for your primary independent variables' analysis (what various types and sources of text data, financial data, monitoring data, as applicable). b) What time frame are using? c) What data frequency?

The data for the primary independent variables’ analysis will be found by key words in financial news or search history and synthesized using phrases of excitement/anxiety. We would like to go back historically to at least 3 economic recessions ago, approximately 1989.

Source Frequency

Principal researchers: Dr. Lucia Alessi (JRC) and Dr. Mikhail Oet (Northeastern University)

FOMC Meeting Minutes Monthly (One per month)

New York Times Articles Monthly (5 per month, if available)

New York Times Comments All for each article

Nexis Uni Monthly (5 per month)

Twitter (if possible) Weekly (50 per week)

Financial Accounts of the United States Annually

3.1.2. a) What data is used for your outcomes (typically should be Flow of Funds choices of individual sectoral agents, e.g. levels/flows of portfolios.) b) What time frame are using? c) What data frequency?

The data for outcomes will be the correlation between independent data and financial and instrumental change (eg, stocks, house pricing, debt).

3.1.3 a) What data is used by your control variables (e.g. macroeconometric conditions)? What time frame are using? c) What data frequency?

We want to control macroeconomic conditions like world events. These events can be found in the GDELT database.

Principal researchers: Dr. Lucia Alessi (JRC) and Dr. Mikhail Oet (Northeastern University)

3.1.4 To be sure that you are identifying stress/sentiment that is distinct from consumer optimism, for example, what are the alternative variables that you want to include to strengthen the case for your latent construct? You may want to investigate appropriate choices among https://fred.stlouisfed.org/searchresults/?st=sentiment and https://www.quandl.com/search?filters=%5B%22Free%22%5D&query=Sentiment. (5 extra points for creativity in finding additional sources e.g. through Snell library).

We could use https://www.aaii.com/sentimentsurvey, also found on https://www.quandl.com/data/AAII- Investor-Sentiment. The AAII investor sentiment survey aims to measure the mood of independent individual investors. We can use this index as a control variable in our analysis, or, we could use the index to compare how sentiment is received by investors in relation to other levels we are using.

Principal researchers: Dr. Lucia Alessi (JRC) and Dr. Mikhail Oet (Northeastern University)

4. Appendix

4.1 Organizations

F.111 U.S. Chartered Depository Institutions

Depository Institution Ticker
1895 Bancorp of Wisconsin Inc BCOW
1st Colonial Bancorp, Inc. FCOB
1st Source Corporation SRCE
AB&T Financial Corporation ABTO
ACNB Corporation ACNB
Alerus Financial Corp ALRS
Allegiance Bancshares Inc ABTX
Amalgamated Bank AMAL
Amerant Bancorp Inc AMTB
American Bank Incorporated AMBK
American Business Bank AMBZ
American National Bankshares Inc. AMNB
American River Bankshares AMRB
Ameris Bancorp ABCB
Ameriserv Financial, Inc. ASRV
Ames National Corporation ATLO
Arrow Financial Corporation AROW

Principal researchers: Dr. Lucia Alessi (JRC) and Dr. Mikhail Oet (Northeastern University)

Associated Banc-Corp ASB Atlantic Capital Bancshares Inc ACBI Atlantic Union Bankshares Corp AUB Auburn National Bancorporation, Inc. AUBN Banc of California Inc BANC Bancfirst Corporation BANF Bancorp 34 Inc BCTF Bancorp of New Jersey Inc BKJ Bancorp, Inc. TBBK Bancorpsouth Bank BXS Bank First Corp BFC Bank of America Corporation BAC Bank of Commerce Holdings BOCH Bank of Hawaii Corporation BOH Bank Of Marin Bancorp BMRC Bank Of New York Mellon Corp. BK Bank Of Princeton BPRN Bank of San Francisco BSFO Bank of South Carolina Corporation BKSC Bank of the James Financial Group, Incorporation BOTJ Bank Ozk OZK Bank7 Corp BSVN Bankwell Financial Group Inc BWFG

Principal researchers: Dr. Lucia Alessi (JRC) and Dr. Mikhail Oet (Northeastern University)

Banner Corporation BANR Bar Harbor Bankshares BHB Bay Banks of Virginia Incorporated BAYK BayCom Corp BCML BDO Unibank Incorporated - ADR BDOUY Berkshire Bancorp Inc. BERK BNCCORP, Inc. BNCC Bogota Financial Corp BSBK BOK Financial Corporation BOKF BOL Bancshares, Inc. BOLB Boston Private Financial Holdings, Inc. BPFH Botetourt Bankshares Inc BORT Bridge Bancorp, Inc BDGE Bridgewater Bancshares Inc BWB Bryn Mawr Bank Corporation BMTC Burke & Herbert Bank & Trust Co BHRB Business First Bancshares Inc BFST Byline Bancorp Inc BY C&F Financial Corporation CFFI Cadence Bancorp CADE California First National Bancorp CFNB Calvin B. Taylor Bankshares, Inc. TYCB Cambridge Bancorporation CATC

Principal researchers: Dr. Lucia Alessi (JRC) and Dr. Mikhail Oet (Northeastern University)

Camden National Corporation CAC Canandaigua National Corporation CNND Capital Bancorp Inc CBNK Capital City Bank Group, Inc. CCBG Capitalsouth Bancorp CAPB Capstar Financial Holdings Inc CSTR Carolina Financial Corp CARO Carolina Trust Bancshares Inc CART Carroll Bancorp Incorporated CROL Carter Bank & Trust CARE Cathay General Bancorp CATY CB Financial Services Inc CBFV CBM Bancorp Inc CBMB CBTX Inc CBTX CCFNB Bancorp, Inc. CCFN Centerstate Bank Corp CSFL Central Pacific Financial Corp. CPF Central Valley Community Bancorp CVCY Century Bancorp, Inc. CNBKA Chemical Financial Corporation TCF Chemung Financial Corporation CHMG Chino Commercial Bancorp CCBC ChoiceOne Financial Services, Inc. COFS

Principal researchers: Dr. Lucia Alessi (JRC) and Dr. Mikhail Oet (Northeastern University)

CIB Marine Bancshares Incorporation CIBH Citigroup Inc. C Citizens & Northern Corporation CZNC Citizens Bancorp Of Virginia, Inc. CZBT Citizens Bancshares Corporation CZBS Citizens Financial Corp. WV CIWV Citizens Financial Group Inc CFG Citizens Financial Services, Inc. CZFS Citizens Holding Company CIZN City Bank CTBK City Holding Co CHCO Civista Bancshares Inc CIVB CNB Corporation CNBW CNB Financial Corporation CCNE Colony Bankcorp, Inc. CBAN Columbia Banking Systems Inc COLB Comerica Incorporated CMA Commerce Bancshares, Inc. CBSH Commercial National Financial Corp CNAF Community Bancorp. CMTV Community Bank System Incorporated CBU Community Bankers Trust Corp ESXB Community Financial Corp TCFC

Principal researchers: Dr. Lucia Alessi (JRC) and Dr. Mikhail Oet (Northeastern University)

Community First Bancorporation CFOK Community Shores Bank Corporation CSHB Community Trust Bancorp, Inc. CTBI Community West Bancshares CWBC ConnectOne Bancorp Inc CNOB Croghan Bancshares, Inc. CHBH Crossfirst Bankshares Inc CFB CSB Bancorp, Inc. CSBB Cullen/Frost Bankers, Inc CFR Customers Bancorp Inc CUBI CVB Financial Corporation CVBF Danske Bank AS - ADR DNKEY Delmar Bancorp DBCP Denmark Bancshares, Inc. DMKBA Dimeco, Inc. DIMC Eagle Bancorp Montana, Incorporation EBMT Eagle Bancorp, Inc. EGBN Eagle Financial Services, Inc. EFSI East West Bancorp Incorporated EWBC Elah Holdings Inc ELLH Elmira Savings Bank ESBK Embassy Bancorp, Incorporation EMYB Emclaire Financial Corp. EMCF

Principal researchers: Dr. Lucia Alessi (JRC) and Dr. Mikhail Oet (Northeastern University)

ENB Financial Corporation ENBP Enterprise Bancorp, Inc. EBTC Enterprise Financial Services Corporation EFSC Equity Bancshares Inc EQBK Esquire Financial Holdings Inc ESQ Evans Bancorp, Inc. EVBN F & M Bank Corp. FMBM F.N.B. Corporation FNB Farmers & Merchants Bancorp FMCB Farmers And Merchants Bank Of Long Beach FMBL Farmers National Banc Corp. FMNB Fauquier Bankshares, Inc. FBSS FB Financial Corp FBK Fentura Financial, Inc. FETM Fidelity D & D Bancorp, Inc. FDBC Fifth Third Bancorp FITB Financial Institutions, Inc. FISI First Bancorp FBNC First Bancorp Inc FNLC First Bancshares Inc FBMS First Bankers Trustshares Inc FBTT First Busey Corporation BUSE First Business Financial Services, Inc. FBIZ

Principal researchers: Dr. Lucia Alessi (JRC) and Dr. Mikhail Oet (Northeastern University)

First Choice Bancorp FCBP First Citizens BancShares Inc FCNCA First Commonwealth Financial Corporation FCF First Community Bankshares Inc FCBC First Community Corp FCCO First Farmers and Merchants Corporation FFMH First Financial Bancorp FFBC First Financial Bankshares, Inc. FFIN First Financial Corporation THFF First Foundation Inc FFWM First Hawaiian Inc FHB First Horizon National Corp FHN First Interstate Bancsystem, Inc. FIBK First Keystone Corporation FKYS First Merchants Corporation FRME First Mid Bancshares Inc FMBH First Midwest Bancorp Incorporated FMBI First National Bank Alaska FBAK First National Corp Virginia FXNC First Nbc Bank Holding Co FNBCQ First Northwest Bancorp FNWB First of Long Island Corp FLIC First Ottawa Bancshares, Inc. FOTB

Principal researchers: Dr. Lucia Alessi (JRC) and Dr. Mikhail Oet (Northeastern University)

First Reliance Bancshares, Inc. FSRL First Republic Bank FRC First Seacoast Bancorp FSEA First United Corporation FUNC First US Bancshares Inc FUSB First Western Financial Inc MYFW FNBH Bancorp, Inc. FNHM FNCB Bancorp Inc FNCB Franklin Financial Network Inc FSB Franklin Financial Services Corporation FRAF Fulton Financial Corporation FULT FVCBankcorp Inc FVCB German American Bancorp, Inc. GABC Glacier Bancorp, Inc. GBCI Glen Burnie Bancorp GLBZ Golden Valley Bank GVYB Grand River Commerce, Incorporation GNRV Great Southern Bancorp, Inc. GSBC Great Western Bancorp Inc GWB Guaranty Bancshares, Inc. GNTY Guaranty Federal Bancshares, Inc. GFED Hancock Whitney Corp HWC Hanmi Financial Corporation HAFC

Principal researchers: Dr. Lucia Alessi (JRC) and Dr. Mikhail Oet (Northeastern University)

Harleysville Financial Corp HARL Harvest Community Bank HCBP Hawthorn Bancshares Inc HWBK HBT Financial Inc HBT Heartland Financial USA, Inc. HTLF Heritage Commerce Corp HTBK Highlands Bankshares Inc HLND Hills Bancorporation HBIA Home Bancshares, Inc. HOMB Hometrust Bancshares Inc HTBI Hope Bancorp Inc HOPE Horizon Bancorp Inc HBNC Horizon Bancorp, Inc. HRRB Howard Bancorp Inc HBMD Huntington Bancshares Incorporated HBAN HV Bancorp Inc HVBC IBERIABANK Corp IBKC Independent Bank Corp INDB Independent Bank Group Inc IBTX International Bancshares Corporation IBOC Investar Holding Corp ISTR Isabella Bank Corporation ISBA Jeffersonville Bancorp JFBC

Principal researchers: Dr. Lucia Alessi (JRC) and Dr. Mikhail Oet (Northeastern University)

Juniata Valley Financial Corp JUVF Katahdin Bankshares Corp KTHN Kentucky Bancshares, Inc. KTYB KeyCorp KEY Killbuck Bancshares Inc KLIB Lakeland Bancorp, Inc. LBAI Lakeland Financial Corporation LKFN Landmark Bancorp, Inc. LARK LCNB Corp LCNB Level One Bancorp Inc LEVL Lifestore Financial Group Inc LSFG Limestone Bancorp Inc LMST Live Oak Bancshares Inc LOB Luther Burbank Corp LBC M&F Bancorp, Inc. MFBP M&T Bank Corporation MTB Macatawa Bank Corporation MCBC Mackinac Financial Corporation MFNC Madison County Financial Inc MCBK MainStreet BankShares Inc MNSB Marlin Business Services Corp. MRLN Mechanics Bank MCHB Medallion Financial Corporation MFIN

Principal researchers: Dr. Lucia Alessi (JRC) and Dr. Mikhail Oet (Northeastern University)

Melrose Bancorp Inc MELR Mercantile Bank Corporation MBWM Merchants Bancorp MBIN Meridian Corp MRBK Metrocity Bankshares Inc MCBS Metropolitan Bank Holding Corp MCB Mid Penn Bancorp, Inc. MPB Middlefield Banc Corp. MBCN Midland States Bancorp Inc MSBI Midwestone Financial Group Inc MOFG Millennium Bankshares Corporation MBVA Mutualfirst Financial, Inc. MFSF National Bank Holdings Corp NBHC National Bankshares Inc NKSH NBT Bancorp Inc. NBTB Neffs Bancorp, Inc. NEFB New Bancorp Inc NWBB New Peoples Bankshares Inc NWPP Nexity Financial Corporation NXTYQ Nicolet Bankshares Inc NCBS Northeast Bank NBN Northern California Bancorp, Incorporation NRLB Northern Trust Corporation NTRS

Principal researchers: Dr. Lucia Alessi (JRC) and Dr. Mikhail Oet (Northeastern University)

Northway Financial, Inc. NWYF Northwest Bancshares Incorporation NWBI Norwood Financial Corp. NWFL Oak Ridge Financial Services Inc BKOR Oak Valley Bancorp OVLY Ohio Valley Banc Corp. OVBC Old National Bancorp ONB Old Point Financial Corporation OPOF Old Second Bancorp, Inc. OSBC Omni Financial Services, Inc. OFSI Open Bank OPBK Optimumbank Holdings, Inc. OPHC Opus Bank OPB Oregon Pacific Bancorp ORPB Origin Bancorp Inc OBNK Orrstown Financial Services, Inc. ORRF Pacific Financial Corporation PFLC Pacific Mercantile Bancorp PMBC Pacific Premier Bancorp, Inc. PPBI Pacwest Bancorporation PACW Park National Corporation PRK Parke Bancorp, Inc. PKBK Parkway Acquisition Corp PKKW

Principal researchers: Dr. Lucia Alessi (JRC) and Dr. Mikhail Oet (Northeastern University)

Parkway Bank PKWY Patriot National Bancorp, Inc. PNBK PCB Bancorp PCB Peapack-Gladstone Financial Corporation PGC Penns Woods Bancorp, Inc. PWOD Peoples Bancorp Inc. PEBO Peoples Bancorp of North Carolina, Inc. PEBK Peoples Bancorp, Inc. PEBC Peoples Financial Corporation PFBX Peoples Financial Services Corp PFIS Peoples Utah Bancorp PUB Pinnacle Bankshares Corporation PPBN Pinnacle Financial Partners, Inc. PNFP Pioneer Bankshares Incorporation PNBI PNC Financial Services Group Incorporated PNC Potomac Bancshares, Inc. PTBS Preferred Bank PFBC Premier Financial Bancorp Inc PFBI Prime Meridian Holding Co PMHG Princeton National Bancorp, Inc. PNBC Professional Holding Corp PFHD Prosperity Bancshares, Inc. PB Provident Bancorp Inc PVBC

Principal researchers: Dr. Lucia Alessi (JRC) and Dr. Mikhail Oet (Northeastern University)

Psb Holdings, Inc. PSBQ QCR Holdings, Inc QCRH QNB Corp. QNBC Rainier Pacific Financial Group, Inc. RPFG RBB Bancorp RBB Red River Bancshares Inc RRBI Regions Financial Corp RF Reliant Bancorp Inc RBNC Renasant Corporation RNST Republic Bancorp, Inc. RBCAA Republic First Bancorp, Inc. FRBK Rhinebeck Bancorp Inc RBKB Richmond Mutual Bancorporation Inc RMBI River Financial Corp RVRF Riverview Financial Corporation RIVE S & T Bancorp, Inc. STBA Sandy Spring Bancorp, Inc. SASR SB Financial Group Inc SBFG SB One Bancorp SBBX Seacoast Banking Corporation of Florida SBCF Select Bancorp Inc SLCT Seneca-Cayuga Bancorp, Inc. SCAY ServisFirst Bancshares, Inc. SFBS

Principal researchers: Dr. Lucia Alessi (JRC) and Dr. Mikhail Oet (Northeastern University)

Severn Bancorp, Inc. SVBI Shore Bancshares, Inc. SHBI Sierra Bancorp BSRR Signature Bank SBNY Silvergate Capital Corp SI Simmons First National Corporation SFNC Skandinaviska Enskilda Banken - ADR SKVKY SmartFinancial Inc SMBK Solera National Bancorp, Incorporation SLRK South Plains Financial Inc SPFI South State Corp SSB Southcrest Financial Group, Inc. SCSG Southeastern Banking Corporation SEBC Southern First Bancshares Inc SFST Southern National Bancorp Of Virginia, Inc. SONA Southside Bancshares, Inc. SBSI Southwest Georgia Financial Corporation SGB Spirit of Texas Bancshares Inc STXB SSB Bancorp Inc SSBP Standard AVB Financial Corp STND Standard Bank Group Limited - ADR SGBLY Sterling Bancorp Inc SBT Stock Yards Bancorp Inc SYBT

Principal researchers: Dr. Lucia Alessi (JRC) and Dr. Mikhail Oet (Northeastern University)

Summit Financial Group, Inc. SMMF Summit State Bank SSBI Sunnyside Bancorp Inc SNNY Suntrust Banks, Inc. STI Surrey Bancorp. SRYB SVB Financial Group SIVB Synovus Financial Corp. SNV Syringa Bancorp SGBP Teb Bancorp Inc TBBA Temecula Valley Bancorp, Inc. TMCV Texas Capital Bancshares, Inc. TCBI Thomasville Bancshares, Inc. THVB Tompkins Financial Corporation. TMP Touchmark Bancshares, Incorporation TMAK Towne Bank TOWN Tri City Bankshares Corporation TRCY Trico Bancshares TCBK Tristate Capital Holdings Inc TSC Triumph Bancorp Inc TBK Truist Financial Corp TFC Trustco Bank Corporation NY TRST Trustmark Corporation TRMK Two Rivers Bancorp TRCB

Principal researchers: Dr. Lucia Alessi (JRC) and Dr. Mikhail Oet (Northeastern University)

Two Rivers Financial Group, Incorporated TRVR U.S. Bancorp USB UMB Financial Corp UMBF Union Bankshares, Inc. UNB United Bancorp, Inc. Ohio UBCP United Bancshares Inc., Ohio UBOH United Bankshares, Inc. UBSI United Community Banks, Inc. UCBI United Security Bancshares UBFO Unity Bancorp, Inc. UNTY University Bancorp Inc UNIB Univest Financial Corp UVSP Uwharrie Capital Corp UWHR Valley National Bancorp VLY Veritex Holdings Inc VBTX Village Bank and Trust Financial Corp. VBFC VSB Bancorp, Inc. NY VSBN Washington Trust Bancorp Inc WASH Webster Financial Corporation WBS Wells Fargo & Company WFC Wesbanco Inc WSBC West Bancorporation Inc WTBA West End Indiana Bancshares Inc WEIN

Principal researchers: Dr. Lucia Alessi (JRC) and Dr. Mikhail Oet (Northeastern University)

West Suburban Bancorp, Inc. WNRP
WestAmerica Bancorp WABC
Western Alliance Bancorporation WAL
Wintrust Financial Corporation WTFC
WSFS Financial Corporation WSFS
Zions Bancorp ZION

F.115 Property-Casualty Insurance Companies

Company Name Ticker
1347 Property Insurance Holdings Inc PIH
ACMAT Corporation ACMT
AFLAC Incorporated AFL
Alleghany Corporation Y
Allstate Corp ALL
Ambac Financial Group Inc. AMBC
American Financial Group Inc. AFG
American International Group Inc. AIG
Amerisafe Inc. AMSF
Arthur J. Gallagher & Co. AJG
Assurant Inc. AIZ
Atlas Financial Holdings Incorporation AFH
Berkshire Hathaway Inc. BRK.A
Blue Water Global Group Inc BLUU

Principal researchers: Dr. Lucia Alessi (JRC) and Dr. Mikhail Oet (Northeastern University)

Brown & Brown Inc. BRO BRP Group Inc BRP Cincinnati Financial Corporation CINF CNA Financial Corporation CNA CNO Financial Group Inc CNO Conifer Holdings Inc CNFR CorVel Corporation CRVL Crawford & Company CRD.A Donegal Group Inc. DGICA Ehealth Inc. EHTH Employers Holdings Inc EIG Equitable Holdings Inc EQH Erie Indemnity Company ERIE Fednat Holding Co FNHC Fidelity National Financial Inc FNF First Acceptance Corporation FACO First American Financial Corporation FAF Gainsco Incorporation GANS Goosehead Insurance Inc GSHD Grand Havana Inc GHAV Hallmark Financial Services Inc. HALL Hanover Insurance Group Inc THG Hartford Financial Services Group Inc HIG

Principal researchers: Dr. Lucia Alessi (JRC) and Dr. Mikhail Oet (Northeastern University)

HCI Group Inc HCI Health Insurance Innovations Inc HIIQ Heritage Insurance Holdings Inc HRTG Hilltop Holdings Inc HTH Horace Mann Educators Corporation HMN Huize Holding Ltd - ADR HUIZ ICC Holdings Inc ICCH InsPro Technologies Corporation ITCC Investors Title Company ITIC Kemper Corporation KMPR Kingstone Companies Incorporation KINS Kinsale Capital Group Inc KNSL Loews Corporation L Markel Corporation MKL Marsh & McLennan Companies Inc. MMC MBIA Inc. MBI Mercury General Corporation MCY MGIC Investment Corporation MTG National General Holdings Corp NGHC NI Holdings Inc NODK Nmi Holdings Inc NMIH Old Republic International Corporation ORI Pacific Ventures Group Inc PACV

Principal researchers: Dr. Lucia Alessi (JRC) and Dr. Mikhail Oet (Northeastern University)

Palomar Holdings Inc PLMR PMI Group Inc PMIR Positive Physicians HoldingsInc PPHI Principal Financial Group Incorporated PFG Proassurance Corporation PRA Progressive Corp PGR Prosight Global Inc PROS Protective Insurance Corp PTVCA QBE Insurance Group Limited - ADR QBIEY Radian Group Incorporated RDN Reinsurance Group of America Inc. RGA RLI Corp. RLI Safety Insurance Group Inc. SAFT Selective Insurance Group Incorporated SIGI State Auto Financial Corp STFC Stewart Information Services Corporation STC Sundance Strategies Inc SUND Travelers Companies Inc TRV Triad Guaranty Inc TGIC Unico American Corporation UNAM United Fire & Casualty Co UFCS United Insurance Holdings Corp UIHC Universal Insurance Holdings Inc UVE

Principal researchers: Dr. Lucia Alessi (JRC) and Dr. Mikhail Oet (Northeastern University)

Unum Group UNM
Voya Financial Inc VOYA
W. R. Berkley Corp WRB

F.116 Life Insurance Companies

Company Name Ticker
American Equity Investment Life Holding Company AEL
American National Insurance Company ANAT
Atlantic American Corporation AAME
Brighthouse Financial Inc BHF
Citizens Financial Corporation CFIN
Citizens, Inc. CIA
Emergent Capital Inc EMGC
FBL Financial Group, Inc. FFG
Federal Life Group Inc FLFG
Galaxy Next Generation Inc GAXY
Genworth Financial, Inc. GNW
Globe Life Inc GL
GWG Holdings Inc GWGH
Independence Holding Company IHC
Kansas City Life Insurance Company KCLI

Principal researchers: Dr. Lucia Alessi (JRC) and Dr. Mikhail Oet (Northeastern University)

Lincoln National Corporation LNC MetLife, Inc. MET National Security Group Inc NSEC National Western Life Group Inc NWLI Primerica, Inc PRI Prudential Financial, Incorporated PRU Sanlam Limited - ADR SLLDY UTG, Inc. UTGN Vericity Inc VERY