Narratives of Fintech in BIS Speeches

EPSA 2025, Madrid

Alfredo Hernandez Sanchez

Vilnius University

2025-06-28

BIS Fintech Speeches


Looking back, the 2014 conference marked the first time we used the term “digital financial inclusion”. It was also the first time most of the organisations present in this room came together to consider the shifting risks posed by the spread of digital transactional platforms designed to bring much needed services to the world’s financially under-served - then estimated at 2.5 billion.

  • Jaime Caruana, BIS, October 2016

FinTech Google Trends

Policy Narratives and Fintech


Narratives are ubiquitous and their political and cognitive importance has been recognized for a long time. Nevertheless, their systematicanalysis in political science and economics has lagged (Schlaufer et al. 2022; Roos and Reccius 2024).


[A]ll narratives are essentially normative, even when the voice of the narrator is well hidden (Patterson and Monroe 1998).

The concepts we use to build theory are themselves narratives, or the symptoms of narratives (Patterson and Monroe 1998).

Research Objectives / Hypotheses


  • To determine which central banks discuss fintech
    • H: countries with less developed traditional finance sectors are more likely to discuss fintech
    • H: countries with more independent financial authorities are more likely to discuss fintech
    • H: countries where fintech is more salient are more likely to discuss fintech
  • To uncover prevalent themes in fintech policy discourse
    • H: countries emphasize themes that address more salient domestic issues (e.g. financial inclusion/less infrastructure)

Workflow

Data Collection


  • Data was collected from the BIS Fintech collection which hosts monetary authority and BIS management speeches and lectures on the topic going back to 2014.

  • The BIS (2019) defines fintech as “technology-enabled innovation in financial services”.

  • The automated webscraping was performed using a headless firefox browser with the RSelenium package on April 30th, 2025, resulting in 675 observations in total.

Webscraping (RSelenium)


library(RSelenium)

human_pause <- function(min = 0.5, max = 1) {
  Sys.sleep(runif(1, min, max))
}

remote_firefox <- rsDriver(
    browser = "firefox",
    chromever = NULL,
    verbose = F,
    port = free_port() 
  )

remDR <- remote_firefox$client

remDR$open()
  
for (i in seq_len(nrow(df_urls))) {
    
  cat("Scraping text for", i, "of", nrow(df_urls), "\n")
    
  url <- df_urls$url[i]
  
  # PDFs are not scraped (yet...)  
  if (str_detect(url, "\\.pdf$")) {
      cat("Skipping PDF link:", url, "\n")
      next
    }
    
  remDR$navigate(url)
  
  human_pause(1.5, 3)
    
  # Extract speech text
  speech_text <- tryCatch({
      body_elem <- remDR$findElement(using = "css selector", 
                                     value = "#cmsContent")
      body_elem$getElementText()[[1]]
    }, error = function(e) {
      cat("No #cmsContent on", url, "\n")
      NA_character_
    })
    
    df_urls$text[i] <- speech_text

  human_pause()

  }

remote_firefox$server$stop()

Document Distribution

BIS Fintech Speeches Map


+−
Leaflet

Fintech Speech Models


Equation 1 corresponds to the cross-sectional LOGIT model.

log(p1−p)=β0+β1ATMsi+β2Bank.Concentrationi+β3Mobile.Utilitiesi+β4CB.Independencei+β5Fintech.Searchi+β6log(GDP)i+β7log(GDP.capita)i+ϵi(1)

Where:

  • Yi,l corresponds to the average propensity towards label c in country i
  • β1ATMsi represents the number of ATMs per 100k people.
  • β2Bank.Concentrationi represents the share of the top 5 banks.
  • β3Mobile.Utilitiesi represents the share of people that have paid utilities with digital means.
  • β4CB.Independencei represents the independence of the central bank.
  • β5Fintech.Searchi represents the Google Trends search rank for fintech.

Which countries talk about Fintech?

Natural Language Inference

  • The process of manual labeling of text segments for training new models is both costly and time consuming (Laurer et al. 2024).
  • Recently, researchers have shifted towards using pre-trained models such as LLMs and BERT-based models (Di Leo et al. 2025; Le Mens and Gallego 2025; Ivacic et al. 2024).
  • Small BERT-based language models are just as efficient at classification tasks while being both deterministic and (mostly) open-source (Burnham et al. 2024).

Labeling Process


  • Classification labels are derived inductively from a qualitative content analysis of a random sample of the corpus.
df_sample <- df %>% 
  sample_n(30)

df_sample %>%
  select(doc_id, text) %>%
  pwalk(~ write_lines(.y, 
                      file.path("output/bis_sample", 
                                paste0(.x, ".txt"))))
  • Each document has its own document id, subsequently each sentence is given its own sentence id keeping in line with tidy data principles.

Zero-Shot Classification


library(reticulate)

transformers <- import("transformers")

classifier <- transformers$pipeline("zero-shot-classification", 
                                    model = "mlburnham/Political_DEBATE_large_v1.0")

labels <- list("financial risk", 
               "economic growth", 
               "economic competition",
               "financial inclusion", 
               "financial innovation",
               "financial efficiency",
               "regulatory cooperation",
               "other")

# Define sentence classifier
classify_sentences_adaptive <- function(df, 
                                        classifier, 
                                        labels) {
  
    # Load required packages (as a block)
    required_packages <- c("dplyr", "tibble", "purrr", "reticulate")
    installed <- installed.packages()
    missing <- setdiff(required_packages, rownames(installed))
    if (length(missing) > 0) install.packages(missing)
    lapply(required_packages, require, character.only = TRUE)
    
    # Clean up and deduplicate labels
    unique_labels <- unique(unlist(labels))
    label_colnames <- make.names(unique_labels, unique = TRUE)
    
    # Function to classify a single sentence
    classify_single <- function(sentence) {
      result <- classifier(sentence, unique_labels)
      scores_r <- unlist(reticulate::py_to_r(result$scores))
      labels_r <- unlist(reticulate::py_to_r(result$labels))
      scores_named <- setNames(as.numeric(scores_r), labels_r)
      score_row <- setNames(
        lapply(unique_labels, function(lbl) scores_named[[lbl]] %||% NA_real_),
        paste0(label_colnames, "_score")
      )
      tibble::as_tibble(score_row)
    }
    
    # Apply classification to each sentence (sequentially)
    results <- purrr::map_dfr(df$sentence, classify_single)
    
    # Bind to original dataframe (returns data frame)
    dplyr::bind_cols(df, results)
}

system.time({
    df_classified <- classify_sentences_adaptive(
      df = df_tokens,
      classifier = classifier,
      labels = labels
    )
})

Classified Sentence Sample


doc_id sentence_id date speaker_name speaker_position iso3c sentence financial.risk_score economic.growth_score economic.competition_score financial.inclusion_score financial.innovation_score financial.efficiency_score regulatory.cooperation_score other_score
r240531a r240531a_102 2024-05-31 Ms Rosanna Costa Governor of the Central Bank of Chile CHL In short, we have reached a very efficient electronic transfer market with instant availability of funds. 0.000 0.000 0.000 0.000 0.309 0.691 0.000 0.000
r211208c r211208c_33 2021-06-09 Ms Nina Stoyanova Deputy Governor and Head of the Banking Department BGR PISA framework aims to cover not only the traditional payment schemes and instruments (e.g. payment cards), but also innovative instruments, such as digital payment tokens backed by assets (e.g. stablecoins). 0.000 0.000 0.000 0.023 0.941 0.036 0.000 0.000
r221017a r221017a_36 2022-10-17 Mr Michael S Barr Vice Chair for Supervision of the Board of Governors of the Federal Reserve System USA For example, it is important for banks to understand some of the heightened liquidity risks they may face from certain types of deposits from crypto-asset companies. 0.980 0.000 0.000 0.000 0.020 0.000 0.000 0.000
r201223o r201223o_44 2020-12-23 Yaron Amir Governor of the Bank of Israel ISR An additional issue that we are promoting is securitization A securitization transaction is one in which a corporation issues bonds whose maturity is secured by expected cash flows. 0.062 0.000 0.000 0.000 0.935 0.001 0.000 0.001
r220719b r220719b_28 2022-07-19 Mr Ravi Menon Managing Director of the Monetary Authority of Singapore SGP Supply chain frictions began to emerge in H2 last year, even before the Russia-Ukraine conflict. 0.000 0.000 1.000 0.000 0.000 0.000 0.000 0.000
r220906e r220906e_131 2022-02-03 Mr Richard Doornbosch President of the Centrale Bank van Curaçao en Sint Maarten CUW Better governance is lesser supervision. 0.079 0.075 0.095 0.075 0.075 0.142 0.104 0.354
r230905f r230905f_29 2023-09-05 Mr Eddie Yue Chief Executive HKG With a lower AB, timely and efficient funding management have become more important than they were when liquidity was plentiful. 0.006 0.000 0.000 0.000 0.000 0.994 0.000 0.000
r220630a r220630a_3 2022-07-01 Prof Yaron Amir Governor of the Bank of Israel ISR Before we dive into a discussion on the activity of payment-transferring money from one person or business to another-I would like to open with a more general question, maybe even semi-philosophical: What, essentially, is money? 0.023 0.024 0.021 0.028 0.258 0.115 0.022 0.510
r220603f r220603f_15 2022-06-03 Prof Joachim Wuermeling Member of the Executive Board DEU In 1928, he came up with chlorofluorocarbons, or CFCs for short. 0.000 0.000 0.000 0.000 0.999 0.000 0.000 0.000
r221017d r221017d_14 2022-10-17 Mr Burkhard Balz Member of the Executive Board DEU If we look at Germany, for instance, the Bundesbank found in a recent survey on payment behaviour that the share of cash transactions has fallen to 58%.2 The trend is obvious: in the previous studies from 2020 and 2017, cash payments accounted for 60% and 74%, respectively. 0.000 0.000 0.000 0.000 0.000 1.000 0.000 0.000
r211109d r211109d_29 2021-11-09 Mr Eddie Yue Chief Executive HKG In other words, it's not just for the good times, but also, perhaps more importantly, for the bad times. 0.095 0.039 0.038 0.029 0.030 0.032 0.028 0.708
r221011b r221011b_48 2022-10-11 Mr Ravi Menon Managing Director SGP Linking faster payment systems solves the cross-border payments problem but not settlements. 0.000 0.000 0.000 0.000 0.858 0.142 0.000 0.000
r240416b r240416b_28 2024-04-16 Mr Eddie Yue Chief Executive HKG We envision this close-knit relationship between the public and private money to continue. 0.000 0.000 0.000 0.987 0.000 0.013 0.000 0.000
r230718l r230718l_63 2023-06-13 Mr Felipe M Medalla Governor PHL Because [in] much of the data that we are looking at, we did not have the kind of exchange rate changes that we were seeing. 0.036 0.042 0.022 0.021 0.021 0.037 0.021 0.802
r211230k r211230k_49 2021-12-30 Mr Benjamin E Diokno Governor PHL Still on speed, the Financial Institutions Strategic Transfer (FIST) Act was signed into law last February and the BSP issued its implementing regulations in May this year-even as banks' bad debts remain manageable to date. 0.318 0.000 0.000 0.000 0.263 0.381 0.039 0.000
r220112c r220112c_62 2022-01-12 Dr Joachim Nagel Member of the Executive Board DEU The Bundesbank must rise to this challenge, and it will do. 0.003 0.000 0.002 0.000 0.000 0.995 0.000 0.000
r211012e r211012e_28 2021-10-12 Mr T Rabi Sankar Deputy Governor IND Put another way, any fintech entity that provides such liquidity services is effectively functioning as a bank and therefore should be subjected to the same legal/regulatory/supervisory regime that a bank is subjected to. 0.022 0.000 0.125 0.000 0.778 0.072 0.003 0.000
r240617c r240617c_64 2024-06-19 Ms Michelle W Bowman Member of the Board of Governors of the Federal Reserve System USA Does innovation meet customer demands for cheaper or new and innovative products? 0.000 0.000 0.233 0.000 0.767 0.000 0.000 0.000
r240306a r240306a_49 2024-03-06 Mr Kazuo Ueda Governor of the Bank of Japan JPN In doing so, it is necessary to consider, with a well-balanced perspective that takes into account both global and domestic circumstances, potential problems that could eventually arise even if we do not face an immediate problem at the moment. 0.013 0.012 0.012 0.011 0.011 0.012 0.012 0.918
r250407p r250407p_2 2025-04-09 Ms Chiara Scotti Deputy Governor ITA I would like to welcome you all to Banca d'Italia and thank the Global Tax Policy Center at the Institute for Austrian and International Tax Law of Vienna University for jointly organizing this conference to discuss the vital link between financial regulation and taxation in the new digital world. 0.000 0.000 0.000 0.000 0.768 0.081 0.150 0.000
sp201010 sp201010_84 2020-10-12 Agustín Carstens General Manager of the BIS BIS Denn was zu Beginn dieses Jahres geschah, ähnelte einem Krieg. 0.070 0.067 0.073 0.071 0.071 0.065 0.068 0.515
r210910c r210910c_36 2021-09-10 Mr Abdul Rasheed Ghaffour Deputy Governor MYS An increase in temperature will lower the resilience of the agroecosystem against pests and pathogens, resulting in lower yields and quality of crops, affecting the supply of raw materials. 0.210 0.118 0.071 0.072 0.071 0.071 0.073 0.314
r211230l r211230l_36 2021-12-30 Mr Benjamin E Diokno Governor of Bangko Sentral ng Pilipinas PHL With physical distancing now a norm, we have seen substantial increase in transactions in PESONet and InstaPay. 0.000 0.437 0.000 0.051 0.512 0.000 0.000 0.000
r230425o r230425o_47 2023-04-25 Mr Felipe M Medalla Governor of Bangko Sentral ng Pilipinas PHL Finally, we are also moving toward an Open Finance Framework. 0.000 0.000 0.000 0.000 0.930 0.070 0.000 0.000
r230622a r230622a_14 2023-06-22 Dr Jorgovanka Tabaković Governor of the National Bank of Serbia SRB We have remained consistent in our determination to maintain the relative stability of the exchange rate not only to keep core inflation at a considerably lower level than headline, but also to preserve overall business, investment and consumer confidence. 0.000 1.000 0.000 0.000 0.000 0.000 0.000 0.000
sp230222 sp230222_44 2023-02-22 Mr Agustín Carstens General Manager of the BIS BIS These foreseeable gains may just be the tip of the iceberg of additional transformations. 0.095 0.684 0.000 0.000 0.221 0.000 0.000 0.000
r210428a r210428a_33 2021-04-28 Mr Benjamin E Diokno Governor of Bangko Sentral ng Pilipinas PHL First, on financial digitalization-.. 0.000 0.000 0.000 0.000 0.977 0.023 0.000 0.000
sp170918 sp170918_123 2017-09-18 Mr Jaime Caruana General Manager CHE We should certainly maintain this attitude and level of commitment in order to deal effectively with the challenges that lie ahead of us. 0.100 0.113 0.100 0.130 0.114 0.112 0.135 0.197
r221104b r221104b_80 2022-11-04 Mr Ravi Menon Managing Director SGP So what is the problem with money that we are trying to solve? 0.000 0.000 0.000 0.000 0.000 1.000 0.000 0.000
r201223i r201223i_26 2020-10-12 François Villeroy de Galhau Governor of the Bank of France and Chairman of the Autorité de contrôle prudentiel et de resolution (ACPR) FRA On top of this, in our day-to-day supervision, we apply the regulation using a proportional approach. 0.045 0.016 0.018 0.017 0.018 0.025 0.061 0.799
r220830d r220830d_56 2022-08-30 Mr Ravi Menon Managing Director SGP Tokenisation enables the monetisation of any tangible or intangible asset. 0.000 0.000 0.000 0.000 0.990 0.010 0.000 0.000
r241111f r241111f_32 2024-11-11 Mr Brad Jones Assistant Governor (Financial System) AUS Foremost here are geopolitical risks, operational risks (including but not limited to cyber risks) and the risks associated with climate change. 0.264 0.051 0.069 0.050 0.052 0.053 0.050 0.411
r220225c r220225c_204 2022-02-25 Dr Johannes Beermann Member of the Executive Board DEU You see the wide range of possible key factors. 0.042 0.020 0.022 0.016 0.018 0.024 0.013 0.844
r210617d r210617d_24 2021-06-17 Margarita Delgado Deputy Governor ESP Moreover, given that these groups may be more vulnerable to fraud and the rising cyber threat, monitoring must be maximised. 0.491 0.011 0.011 0.012 0.012 0.011 0.053 0.398
r240531a r240531a_44 2024-05-31 Ms Rosanna Costa Governor of the Central Bank of Chile CHL However, there was a need to increase transparency and promote financial stability by mitigating systemic risk. 0.520 0.000 0.000 0.000 0.000 0.480 0.000 0.000
r210323d r210323d_69 2021-03-23 Mr Howard Lee Deputy Chief Executive HKG Conclusion I have highlighted just three areas, namely payment and CBDC, use of data, and digitalization of our own organization, as some of the key challenges many of us are facing. 0.000 0.000 0.000 0.000 1.000 0.000 0.000 0.000
r210607a r210607a_12 2021-06-07 Ignazio Visco Governor of the Bank of Italy ITA The prompt provision and exceptional quantity of liquidity provided by the central banks staved off tensions in the markets and restored broadly accommodative financial conditions. 0.274 0.000 0.000 0.000 0.000 0.726 0.000 0.000
r221116c r221116c_113 2022-11-16 Burkhard Balz Member of the Executive Board DEU As I have already indicated, we are well aware of this. 0.234 0.069 0.196 0.064 0.067 0.121 0.057 0.193
sp201217 sp201217_17 2020-12-17 Mr Benoît Cœuré Head of the BIS Innovation Hub BIS 3 See D Rees, "What comes next? 0.072 0.100 0.099 0.055 0.149 0.052 0.049 0.424
r230308l r230308l_8 2023-03-08 Mr Fabio Panetta Member of the Executive Board ECB We simply cannot afford to fall behind the curve: cybersecurity must be the backbone of digital finance. 0.018 0.000 0.000 0.000 0.982 0.000 0.000 0.000
r211202k r211202k_16 2021-12-02 François Villeroy de Galhau Governor of the Bank of France and Chairman of the Autorité de contrôle prudentiel et de resolution (ACPR) FRA The second generation, i.e. stablecoins – with mechanisms to stabilise their value against sovereign currencies – aims to provide a more comprehensive range of services with a global reach. (iii) Lastly, the emergence of decentralised market infrastructures: new technologies tend to reduce the use of financial intermediaries or centralised systems, often developed by central banks, such as TARGET2, the Real-Time Gross Settlement (RTGS) system for euro payments, or the TARGET2-Securities settlement platform, both of which were developed and are operated by the Eurosystem. 0.000 0.000 0.000 0.000 0.741 0.259 0.000 0.000
r240813c r240813c_94 2024-08-14 Dr Michael Debabrata Patra Deputy Governor IND I wish the Conference all success. 0.075 0.077 0.074 0.075 0.101 0.074 0.090 0.434
r230907e r230907e_95 2023-09-07 Prof Joachim Wuermeling Member of the Executive Board DEU While the past two years have been tough in terms of raising funds, German and European FinTech companies still have a lot of potential yet to be fulfilled. 0.000 0.000 0.000 0.000 0.997 0.003 0.000 0.000
r241126w r241126w_54 2024-11-29 Gabriel Makhlouf Governor of the Central Bank of Ireland IRL Such technologies could provide greater access and choice to people across the world to financial services. 0.000 0.000 0.050 0.097 0.800 0.053 0.000 0.000
r241104k r241104k_12 2024-11-04 Mr Leong Sing Chiong Deputy Managing Director (Markets & Development) of the Monetary Authority of Singapore SGP The beneficiary within Ant International's network can receive its funds in its domiciled currency, for instance US Dollar, in the form of a tokenised deposit. 0.000 0.000 0.000 0.027 0.972 0.001 0.000 0.000
r201204i r201204i_3 2020-12-04 Mr Adnan Zaylani Mohamad Zahid Assistant Governor MYS As a prestigious research academy, it needs to produce prestigious and breakthrough research, and it needs to have events like this dialogue to further advance the knowledge and exchange of ideas and the cause of Islamic Finance. 0.000 0.000 0.000 0.000 1.000 0.000 0.000 0.000
r211123d r211123d_8 2021-11-23 Christine Lagarde President of the European Central Bank DEU This role reversal is partly attributable to the contribution of European banking supervision. 0.000 0.000 0.010 0.000 0.000 0.009 0.980 0.000
r220811e r220811e_3 2022-05-26 Mr Benjamin E Diokno Governor of Bangko Sentral ng Pilipinas PHL The Philippines continues to beat the odds. 0.000 1.000 0.000 0.000 0.000 0.000 0.000 0.000
r231227f r231227f_56 2023-09-28 Mr Constantinos Herodotou Governor of the Central Bank of Cyprus CYP The aim is to reduce risks to the financial system, stemming from money laundering, terrorist financing, bypassing of sanctions, cyber-attacks, fraud or leakages of personal information. 0.996 0.000 0.000 0.000 0.000 0.000 0.004 0.000
r240828p r240828p_88 2024-08-29 Shaktikanta Das Governor of the Reserve Bank of India IND Recognising the same, international bodies such as the Organisation for Economic Cooperation and Development (OECD) have outlined core principles governing AI, which include inclusive growth, respect for the rule of law and human rights, transparency and explainability, robustness and safety, and accountability. 0.000 0.978 0.000 0.000 0.001 0.000 0.021 0.000

Label Distribution (Checks)

Cross-sectional Analysis

  • The dependent variable for the cross-sectional analysis is the average sentence classification (2020-2025) score aggregated at the country level.
  • Independent variables are political-economy indicators for 2020. When data for 2020 is missing, 2021 is imputed.
cs_df <- classified_df %>%
  filter(date >= ymd("2019-12-31")) %>% 
  group_by(iso3c) %>%
  summarise(across(ends_with("_score"),
                   list(mean = ~mean(.x, na.rm = TRUE)),
                   .names = "{.col}_{.fn}"))

Label Models


Equation 2 corresponds to the cross-sectional OLS model.

Yi,c=β0+β1ATMsi+β2Bank.Concentrationi+β3Mobile.Utilitiesi+β4CB.Independencei+β5log(GDP)i+β6log(GDP.capita)i+ϵi(2)

Where:

  • Yi,l corresponds to the average propensity towards label c in country i
  • β1ATMsi represents the number of ATMs per 100k people.
  • β2Bank.Concentrationi represents the share of the top 5 banks.
  • β3Mobile.Utilitiesi represents the share of people that have paid utilities with digital means.
  • β4CB.Independencei represents the independence of the central bank.

Which countries prioritize which labels?

Limitations

  • The cross sectional analysis obviates temporal dynamics.
  • The “narratives” approch requires some refinement.
  • The zero-shot classification must be cross-checked with other models (BERT-based) and other approaches (few-shot classification, LLMs).
  • The corpus is relatively small (n = 675) and (policy) narratives exist in broader fora.

Thank you for your attention!

About the FIRSA Project

Disclaimer:

This project has received funding from the European Union Marie Skłodowska-Curie Postdoctoral Fellowships / ERA Fellowships action under grant agreement No. 101180601 under the title: Understanding FinTech Regulatory Sandbox Development in Europe (FIRSA).

Learn more at the project website.

Logo

Appendix

Data Cleaning (LLM)


gpt_get_structured_output <- function(desc_text, api_key) {
  
    safe_desc <- gsub("\\\\", "\\\\\\\\", desc_text)
    safe_desc <- gsub("\"", "\\\\\"", safe_desc)
    safe_desc <- gsub("\n", "\\\\n", safe_desc)
    safe_key <- gsub("\\\\", "\\\\\\\\", api_key)
    safe_key <- gsub("\"", "\\\\\"", safe_key)
    
    sys_prompt <- paste0(
      "Extract structured metadata from financial speech descriptions. ",
      "If multiple speakers are mentioned, infer the main speaker based on seniority. ",
      "Extract main speaker's position in organization. ",
      "Extract or infer the speaker’s country from their position or affiliation. ",
      "Extract the speech city. ",
      "Extract or infer the speech country from the city or context. ",
      "Extract the speech date, if it is two or more days, get the last. ",
      "Extract if multiple speakers are identified, return true or false. ",
      "Return only the extracted data as key-value pairs, one per line, starting directly with `speaker_name:`. ",
      "If a field is unknown, write `unknown`. Do not include any explanation, header, or formatting. Only return plain structured text."
    )
    
    py_run_string(paste0(
      "import openai\n",
      "client = openai.OpenAI(api_key=\"", safe_key, "\")\n",
      "response = client.chat.completions.create(\n",
      "  model='gpt-3.5-turbo',\n",
      "  temperature=0,\n",
      "  messages=[\n",
      "    {'role': 'system', 'content': \"", sys_prompt, "\"},\n",
      "    {'role': 'user', 'content': \"", safe_desc, "\"}\n",
      "  ]\n",
      ")\n",
      "result = response.choices[0].message.content\n"
    ))
    
    py$result 
  }

Model Correlogram

References


BIS. 2019. “Innovation, Fintech and AI.” September 18, 2019. https://www.bis.org/topic/fintech.htm.
Burnham, Michael, Kayla Kahn, Ryan Yank Wang, and Rachel X. Peng. 2024. “Political DEBATE: Efficient Zero-Shot and Few-Shot Classifiers for Political Text.” arXiv. https://doi.org/10.48550/ARXIV.2409.02078.
Di Leo, Riccardo, Chen Zeng, Elias Dinas, and Reda Tamtam. 2025. “Mapping (a)ideology: A Taxonomy of European Parties Using Generative LLMs as Zero-Shot Learners.” Political Analysis, April, 1–8. https://doi.org/10.1017/pan.2025.7.
Ivacic, Nikola, Matthew Purver, Fabienne Lind, Hajo Boomgaarden, Veronika Bajt, and Senja Pollak. 2024. “Comparing News Framing of Migration Crises Using Zero-Shot Classification.” In Proceedings of the 1st Workshop on Reference, Frames and Perspectives (LREC 2024 Workshop), 18–27.
Laurer, Moritz, Wouter van Atteveldt, Andreu Casas, and Kasper Welbers. 2024. “Less Annotating, More Classifying: Addressing the Data Scarcity Issue of Supervised Machine Learning with Deep Transfer Learning and BERT-NLI.” Political Analysis 32 (1): 1–16. https://doi.org/10.1017/pan.2022.15.
Le Mens, Gaël, and Aina Gallego. 2025. “Positioning Political Texts with Large Language Models by Asking and Averaging.” Political Analysis, January, 1–9. https://doi.org/10.1017/pan.2024.29.
Patterson, Molly, and Kristen Renwick Monroe. 1998. “NARRATIVE IN POLITICAL SCIENCE.” Annual Review of Political Science 1 (1): 315–31. https://doi.org/10.1146/annurev.polisci.1.1.315.
Roos, Michael, and Matthias Reccius. 2024. “Narratives in Economics.” Journal of Economic Surveys 38 (2): 303–41. https://doi.org/10.1111/joes.12576.
Schlaufer, Caroline, Johanna Kuenzler, Michael D. Jones, and Elizabeth A. Shanahan. 2022. “The Narrative Policy Framework: A Traveler’s Guide to Policy Stories.” Politische Vierteljahresschrift 63 (2): 249–73. https://doi.org/10.1007/s11615-022-00379-6.
Narratives of Fintech in BIS Speeches EPSA 2025, Madrid Alfredo Hernandez Sanchez Vilnius University 2025-06-28

  1. Slides

  2. Tools

  3. Close
  • Narratives of Fintech in BIS Speeches
  • BIS Fintech Speeches
  • FinTech Google Trends
  • Policy Narratives and Fintech
  • Research Objectives / Hypotheses
  • Workflow
  • Data Collection
  • Webscraping (RSelenium)
  • Document Distribution
  • BIS Fintech Speeches Map
  • Fintech Speech Models
  • Which countries talk about Fintech?
  • Natural Language Inference
  • Labeling Process
  • Zero-Shot Classification
  • Classified Sentence Sample
  • Label Distribution (Checks)
  • Cross-sectional Analysis
  • Label Models
  • Which countries prioritize which labels?
  • Limitations
  • Thank you for your attention!
  • .f-outer-container...
  • Appendix
  • Data Cleaning (LLM)
  • Model Correlogram
  • References
  • f Fullscreen
  • s Speaker View
  • o Slide Overview
  • e PDF Export Mode
  • r Scroll View Mode
  • ? Keyboard Help