This calculator compares your quoted rate to weekly historical averages, simulates mean-reverting paths to estimate refi odds, and suggests whether paying points is worth it versus waiting to refinance.
Enter your quoted rate and run analysis. Results will show market zone, estimated refi opportunity, and a directional suggestion — alongside confidence intervals where the API provides them.
Historical context uses Freddie Mac PMMS from FRED.Compare a borrower's quoted rate to historical Freddie Mac survey rates, run a statistical simulation over the next several years, and get market context: a market zone, an estimated chance that rates could fall enough to make refinancing worthwhile, and a directional suggestion on points vs lender credit. Share this as context with borrowers — not as a guarantee.
When historical volatility is high, simulated refi probabilities can move in both directions — the model reflects that environment, not only the latest trend.
Technical: The borrower's current quoted mortgage rate in percent (e.g. 6.0 = 6%). This is the starting point r0 for all Monte Carlo simulation paths in the Vasicek model: r(t+1) = r(t) + kappa(theta - r(t))dt + sigma*epsilon.
Intuitive: "What rate did the borrower lock in (or get quoted)?" Every simulation path starts here and wanders from this point. It's the baseline everything is measured against — "will rates drop enough below this number to make refinancing worth it?"
Technical: An optional array of historical rate values (oldest-first, in %). When omitted, the server pulls the Freddie Mac PMMS 30-year weekly series from FRED. Requires minimum 3 data points. Used to derive three key model parameters: the long-run mean (theta), the volatility of rate changes (sigma), and optionally kappa via OLS regression.
Intuitive: "What past rate data should we learn from?" By default it uses real-world Freddie Mac weekly averages — decades of actual mortgage rate history. You'd override this if you wanted to analyze a specific product, a different loan type, or a custom scenario. The model looks at this history to figure out: "what's normal?", "how jumpy are rates?", and "how fast do they snap back?"
Technical: The forward time horizon in years for the Monte Carlo simulation. Each path is simulated with monthly time steps (dt = 1/12) for this many years. Minimum 1 year.
Intuitive: "How far into the future are we looking for a refinance opportunity?" 5 years means: "what are the odds rates drop enough to refi sometime in the next 5 years?" A shorter horizon is more conservative ("I need rates to drop soon"), a longer one gives more chances for rates to dip low enough.
Technical: The basis-point drop below current_rate that constitutes a refinancing opportunity. 75 means the borrower needs rates to fall 0.75% below their current rate. A Monte Carlo path "triggers" a refi event the first time r(t) <= current_rate - trigger/100.
Intuitive: "How big a rate drop does the borrower need before refinancing makes financial sense?" This accounts for closing costs — a tiny rate drop isn't worth the fees. 75 bps (0.75%) is a common rule of thumb. A borrower with a large loan balance might set this lower (even a small drop saves a lot), while someone with a small loan might need a bigger drop to justify the costs.
Technical: The number of independent Monte Carlo simulation paths. Each path is a random walk of monthly rate values drawn from the Vasicek process. The refi probability is estimated as (paths that trigger) / (total paths), with a Wilson score 95% confidence interval.
Intuitive: "How many times do we roll the dice?" More paths = more precise probability estimates, but slower computation. 1,000 gives you a reasonable estimate. 5,000 gives you tighter confidence intervals. Think of it like polling — more respondents means a smaller margin of error. For most users, the default is fine.
Technical: The mean-reversion speed parameter (yr^-1) in the Vasicek model. Governs the drift term kappa(theta - r)dt — how strongly rates are pulled toward the long-run mean theta. Half-life of a deviation is approximately ln(2)/kappa. Ignored when auto_kappa is true.
Intuitive: "When rates get weird, how long do they stay weird?" At the default 0.15, a rate spike has a half-life of about 4.6 years — it takes that long for half the deviation to correct. At kappa=1.0, the half-life is about 8 months — spikes are short blips. At kappa=0.05, it is about 14 years — rates can stay elevated for a very long time. Higher kappa means "the market always corrects quickly." Lower kappa means "rate environments can persist for years."
Technical: When true, kappa is estimated from the historical rate series via OLS regression on the discrete Vasicek equation: delta_r(t) = a + b*r(t) + epsilon, where kappa = -b*12 (annualized from monthly). Clamped to [0.01, 2.0]. Falls back to 0.15 if the series is too short. Takes precedence over the manual kappa field.
Intuitive: "Should the model figure out mean-reversion from the data itself?" Instead of you guessing how fast rates snap back, it measures how fast they actually snapped back historically. This is data-driven rather than assumption-driven. The tradeoff: historical mean-reversion might not reflect the future (e.g., if the Fed changes its policy approach). Manual kappa lets you inject your own view of the market.
Technical: Additional refi trigger thresholds (in basis points) to include in the sensitivity grid. Combined with the default refi_trigger_bps value and the sensitivity horizons to produce a matrix of Monte Carlo results. Each cell in the grid runs 500 paths. Total grid capped at 20 cells.
Intuitive: "What if the borrower needs a different rate drop to justify refinancing?" This lets you see probabilities across multiple scenarios side-by-side. For example, pass [50, 75, 100, 125] to answer: "What are the odds of a half-point drop? Three-quarters? A full point? A point and a quarter?" Great for showing borrowers a range of outcomes instead of a single number.
Technical: Additional time horizons (in years) for the sensitivity grid. Combined with the trigger values to produce the full matrix. Each cell runs 500 independent Monte Carlo paths.
Intuitive: "What if the borrower is willing to wait longer — or needs it sooner?" Pass [1, 3, 5, 7] to show: "Here's the probability of a refi opportunity within 1 year, 3 years, 5 years, and 7 years." Paired with multiple triggers, you get a complete picture — a grid the loan officer can walk through with the borrower: "If you can wait 5 years and need a 75bp drop, there's an X% chance."
Together, these inputs define a question:
"Given where rates are today (current_rate), what we know about how rates have behaved (historical_rates), how quickly they revert to normal (kappa/auto_kappa), how far into the future we're looking (simulation_years), and how big a drop the borrower needs (refi_trigger_bps) — what's the probability a refinance opportunity appears?"
The sensitivity grid parameters (sensitivity_triggers_bps + sensitivity_horizons_years) then expand that single question into a matrix of scenarios the loan officer can present to clients.
