引導式測驗 · 中等 等級
無銀行帳戶者與微型金融的 AI 信用測驗
探討替代數據如何影響信貸獲取,同時強調覆蓋範圍、同意、代理商和消費者保護限制。
問題 1 的 8
為什麼借款人的傳統信用檔案可能很弱?
預覽每個問題8 問題
當您參加測驗時,就會出現答案和解釋。
- 為什麼借款人的傳統信用檔案可能很弱?
- A recent score decline within a long formal loan history
- A high debt-to-income ratio across several reported loans
- A large balance on many established credit accounts
- Limited formal borrowing or barriers to account access
- 使用更多數據是否會自動產生更公平或更實惠的信貸?
- Yes, more inputs always improve a model
- No; data coverage, accuracy, consent and proxy effects matter
- Only for microfinance
- Yes, if the data are digital
- 世界銀行 2025 年全球金融指數衡量什麼?
- Only default rates on microfinance loans
- Exact credit-score model accuracy across countries
- Financial-service access/use and digital connectivity across economies
- Only bank profitability in high-income countries
- 當只有經過批准的借款人觀察到還款結果時,會出現哪種訓練資料問題?
- Only previously approved applicants may have observed repayment outcomes
- Everyone receives the same score
- Data cannot be separated by time
- The model refuses to use a spreadsheet
- 為什麼要檢查替代資料來源中缺少哪些人?
- Missing applicants should automatically receive the highest-risk score
- A different borrower’s transactions can replace the missing record
- Coverage gaps may exclude applicants with limited connectivity or data access
- An evaluation can ignore people absent from the source
- What should a provider record about data features?
- Only the vendor’s stated accuracy claim
- Only whether the source is digital
- Only the final score after the decision
- Provenance, purpose, consent, missingness and update frequency
- Why can shared-device use weaken an alternative-data credit signal?
- The digital activity may not belong to one borrower or represent that borrower’s repayment capacity
- Every shared phone automatically produces a valid credit history
- Shared-device activity proves that a borrower has defaulted
- A phone identifier removes the need for consent or correction
- What else contributes to financial inclusion besides approval rates?
- Only the number of applications accepted
- Affordability, loan terms, recourse and product design
- Only a model’s discrimination metric
- Only the quantity of alternative-data features