Credit score Study Guide

📖 Core Concepts Credit Score – A numeric value (e.g., 300‑850 in the U.S.) produced by statistical models that reflects the probability a borrower will be 90 days past‑due or worse within a set horizon (usually 24 months). Default Probability – The core predictive target; higher scores ⇒ lower default risk. Scorecard – A set of weighted variables (payment history, debt, personal data, etc.) summed to produce the final score. Logistic Regression – The most common technique for binary outcomes (default vs non‑default); it estimates the log‑odds of default and converts them to a probability. Alternative Modelling Methods – Multivariate Adaptive Regression Splines (MARS), Classification & Regression Trees (CART), Chi‑square Automatic Interaction Detector (CHAID), Random Forests. FICO – Fair Isaac Corporation’s flagship scoring model; multiple versions (e.g., Model II, Classic04, industry‑specific) are used across lenders. Consumer Rights (U.S./Australia) – Free annual credit‑report copy from each bureau; free score only under specific denial circumstances. Dispute Process – Under the Fair Credit Reporting Act (FCRA) bureaus must investigate disputed items within 45 days (30 days for paid‑for reports). --- 📌 Must Remember Primary Uses – Approval decisions, credit‑limit setting, collections scoring, pre‑approval of additional credit. Major U.S. Bureaus – Experian, TransUnion, Equifax. FICO Score Range – $300$ – $850$. Mortgage Scoring Practice (U.S.) – Beacon 5.0 (Equifax), FICO Model II (Experian), Classic04 (TransUnion). Alternative Score – VantageScore 3.0 (offered by TransUnion). Free Report Frequency – One per bureau every 12 months (AnnualCreditReport.com). Free Score Eligibility – When credit, insurance, or loan is denied due to the score (Wall Street Reform Act, 22 July 2010). Predictive Horizon – FICO predicts 90‑day+ delinquency within the next 24 months. --- 🔄 Key Processes Scorecard Development (Logistic Regression) Collect historical credit data (payment history, debt, relationships, personal updates). Split into training / validation sets. Fit logistic model: $\log\left(\frac{p}{1-p}\right)=\beta0+\beta1X1+\dots+\betakXk$, where $p$ = probability of default. Convert estimated probabilities to a numeric score (e.g., linear transformation to 300‑850). Validate using out‑of‑sample AUC, KS, or Gini. Consumer Dispute Workflow (U.S.) Consumer identifies an error on the report. Files dispute with the bureau (online, mail, phone). Bureau acknowledges and begins investigation (≤ 45 days). Bureau contacts data furnisher, collects evidence. Decision communicated; if error confirmed, report is corrected; consumer receives updated copy. Obtaining a Free Annual Credit Report Visit AnnualCreditReport.com. Request one report from each of the three bureaus (choose any order). Verify identity (SSN, DOB, address). Download/print the report; review for inaccuracies. --- 🔍 Key Comparisons Logistic Regression vs. Random Forests Logistic: Interpretable coefficients, easy to translate into scorecards. Random Forest: Higher predictive power on complex non‑linear interactions, but less transparent. Private Scores vs. Business Scores (Denmark) Private: Predict individual default risk. Business: Predict corporate bankruptcy risk. FICO vs. VantageScore FICO: Dominant, many versions, industry‑specific adjustments. VantageScore: Single version (3.0), created jointly by the three bureaus; alternative for lenders seeking a different model. Free Credit Report vs. Free Credit Score Report: Always free annually; contains account history, inquiries, public records. Score: Not included for free unless denial triggers a mandatory provision. --- ⚠️ Common Misunderstandings “Free credit report = free credit score.” The annual free report does not include the numeric score. “All FICO scores are identical.” There are 29+ versions; scores differ by loan type and industry weighting. “Higher score guarantees loan approval.” Scores are a major factor, but lenders also consider income, debt‑to‑income ratio, and policy thresholds (often undisclosed). “Dispute resolution is instant.” Legally, bureaus have up to 45 days; some investigations take the full period. --- 🧠 Mental Models / Intuition Score ≈ “Risk Meter” – Think of the score as a dial: 300 = “high chance of default,” 850 = “very low chance.” Logistic Curve – As risk factors increase, the probability of default rises steeply near the middle of the score range, flattening at the extremes. “Layered Scores” – Different loan products (revolving vs. mortgage) use tailored versions of the same underlying model, explaining why you can see multiple FICO numbers on a credit report. --- 🚩 Exceptions & Edge Cases Industry‑Specific FICO Versions – Automotive, bank‑card, and mortgage scores adjust the weighting of variables (e.g., payment history may matter more for mortgages). Multiple Scores per Consumer – A borrower may have separate scores for revolving credit, mortgage, and auto loans; they are not interchangeable. International Variations – Denmark uses both private and business scores; the UK does not require lenders to disclose minimum scores. --- 📍 When to Use Which Choosing a Modelling Technique Logistic Regression: When interpretability and regulatory transparency are required (e.g., credit‑card scorecards). Random Forest / CART: When you have large, high‑dimensional data and need higher predictive accuracy, accepting a “black‑box” model. Selecting a Scoring Model for a Loan Mortgage: Use the three‑bureau approach (Beacon 5.0, FICO II, Classic04). Auto/Bank‑Card: Use the industry‑specific FICO version (e.g., FICO Auto). Alternative Lending: Consider VantageScore 3.0 if the lender wants a non‑FICO benchmark. When to Cite Consumer Rights In exam questions about data access: cite the annual free report right and the conditional free‑score provision (Wall Street Reform Act, 2010). --- 👀 Patterns to Recognize Score Range → Risk Direction – Any question showing a numeric score within 300‑850 can be mapped to a risk level: lower → higher default probability. “Three‑Score Mortgage” – Whenever mortgage underwriting is mentioned, expect the trio (Beacon 5.0, FICO II, Classic04). Model Mention + “Binary Outcome” – Usually points to logistic regression as the intended technique. “Consumer may obtain free copy once per year” – Signals Australian or U.S. consumer‑rights context. --- 🗂️ Exam Traps Distractor: “Free annual credit score is provided.” – Wrong; only the report is free. Distractor: “All FICO scores range 0‑1000.” – Incorrect; the standard range is $300$–$850$. Distractor: “Logistic regression is the only possible modelling method.” – False; MARS, CART, CHAID, Random Forests are also viable. Distractor: “Credit bureaus must resolve disputes in 30 days.” – Only for paid‑for reports; the standard is 45 days. Distractor: “UK lenders must disclose minimum approval scores.” – Not required; disclosure practices differ. ---