Virtual Economist:
AI-Powered Mortgage Rate and Lock Volume Forecasting
Executive Summary
Virtual Economist is a machine-learning and AI-powered forecasting tool that produces 12-month mortgage rate and lock volume outlooks using public economic indicators and aggregated Optimal Blue mortgage market data. Rather than relying on recent trends alone, its predictive machine-learning models evaluate how monetary policy, inflation expectations, market volatility, credit conditions, and mortgage-market activity may influence future outcomes.
The component-based architecture forecasts the 10-Year Treasury, MBS spread, and primary-secondary spread (PSS) separately and combines them into an Optimal Blue Mortgage Market Indices (OBMMI™) mortgage-rate outlook. That rate forecast then becomes an input to the Lock Volume model, creating an integrated view of both mortgage rates and expected market activity.
Rolling out-of-sample testing indicates that the selected Virtual Economist methodology produced lower OBMMI forecast error than the trend, mean-reversion, and persistence benchmarks evaluated in this paper. Results should be interpreted as decision-support information under documented baseline assumptions, not as guaranteed outcomes or business recommendations.
What AI-powered means: Underlying numerical forecasts are produced by predictive machine-learning and time-series models. Generative AI supports the conversational experience, but it does not independently create the numerical forecasts.
Why Historical Forecasting Baselines Can Fall Short
Mortgage lenders consider future interest rates and loan volumes when evaluating budgets, capacity, risk, and business scenarios. Common forecasting baselines often extend a recent trend, assume rates will return toward a long-term average, or carry the current value forward as the next period's estimate. These approaches provide useful reference points, but they may not fully reflect changing monetary policy, inflation expectations, market volatility, credit conditions, and mortgage market activity.
Virtual Economist complements these historical baselines by using machine learning to evaluate a broader set of economic and mortgage-market variables. In the validation tests described in this paper, this approach produced more accurate OBMMI forecasts than the trend, mean-reversion, and persistence methods evaluated.
A Transparent, Component-Based Architecture
Virtual Economist builds its mortgage rate forecast from the ground up. It models each major economic component separately so users can understand which part of the market is contributing to a forecast movement:
| Component | What it captures |
|---|---|
| 10-Year Treasury | The macroeconomic base rate associated with monetary policy, inflation, and growth expectations. |
| MBS Spread | The additional yield associated with mortgage-backed securities relative to the 10-year Treasury. |
| Primary-Secondary Spread | The difference between secondary-market mortgage yields and primary mortgage rates. |
| Lock Volume | Expected monthly mortgage lock activity, using the OBMMI rate forecast as an input. |
This decomposition provides more than a single forecast number. It gives users a framework for understanding whether an expected rate movement originates primarily in the Treasury market, the mortgage-backed securities market, or the primary-secondary spread.
Data Used to Train and Operate the Models
Virtual Economist uses documented public economic indicators and aggregated Optimal Blue mortgage-market data. Historical observations are used to train and evaluate the models, while current and forward-looking assumptions are used to produce the baseline forecast.
Public economic and Optimal Blue mortgage-market data
| Model | Illustrative inputs |
|---|---|
| 10-Year Treasury | Federal Funds Rate, unemployment, real disposable personal income, market volatility, Brent crude oil, 10-year breakeven inflation, economic-policy indicators, and GDP growth. |
| MBS Spread | Corporate credit spreads, interest-rate volatility, Federal Reserve MBS holdings, and mortgage application indices. |
| Primary-Secondary Spread | 2-year and 10-year Treasury rates, current-coupon yield, mortgage-market activity, and profitability-related metrics. |
How the Baseline Forecast Is Constructed
Before the machine learning models generate mortgage rate and lock volume projections, Virtual Economist establishes a baseline economic scenario for the forecast period. The baseline represents a central path using information available when the forecast is produced. It is not based on a single economist's opinion and does not assume that every input remains unchanged.
For the mortgage-rate forecast, Virtual Economist combines three types of forward-looking information:
Market-implied expectations
Where liquid markets express expectations about future conditions, those signals are incorporated into the baseline. Examples include Federal Funds expectations derived from CME FedWatch and Brent crude assumptions derived from the ICE futures curve. These market-based inputs help the forecast respond to information already reflected in traded prices.
Forecasts from recognized public and industry sources
For variables not directly represented by a suitable market curve, the baseline uses published projections. The documented process uses sources such as the Congressional Budget Office or Federal Reserve for unemployment, GDPNow for current-quarter economic growth, Congressional Budget Office projections for the later forecast horizon, Federal Reserve Bank of Atlanta Business Inflation Expectations, and Mortgage Bankers Association forecasts for mortgage-market activity.
Documented assumptions and transformations
Some variables do not have a reliable published 12-month forecast. For these inputs, the baseline uses transparent, repeatable assumptions, such as extending a recent value, applying an established growth assumption, interpolating between near-term and longer-term projections, or applying a documented relationship to another forecast variable.
The resulting variables are supplied to the component models that forecast the 10-Year Treasury, MBS spread, and primary-secondary spread. Their outputs are then combined:
The baseline OBMMI forecast becomes a direct input to the Lock Volume model. The volume model evaluates that rate path together with historical Optimal Blue lock activity and modeled time-series patterns to estimate monthly lock volume. This creates an integrated forecasting chain:
Because the baseline is explicit, users can distinguish between two questions:
- What does the model expect under the current baseline?
- How could the outlook change if an economic assumption changes?
Virtual Economist's scenario analysis capability addresses the second question by modifying selected baseline assumptions and recalculating their modeled effect on mortgage rates and lock volume. A scenario is a structured conditional analysis, not a prediction that the assumed event will occur.
Reading the forecast: The baseline is a conditional outlook based on the economic assumptions available when it is generated. If policy expectations, inflation, energy prices, volatility, or other inputs change, the forecast may change as well.
What Users Receive
Virtual Economist provides:
Monthly forecasts
Monthly mortgage rate and lock volume forecasts grounded in the documented baseline, with updated market assumptions.
Component-level views
Views of the 10-Year Treasury, MBS spread, and primary-secondary spread.
Forecast explanations
Explanations of the factors contributing to forecast movements.
Scenario analysis
Structured “what-if” scenario analysis using adjusted economic assumptions.
Conversational exploration
A conversational experience for exploring forecasts and scenarios.
These outputs can support scenario evaluation, budgeting, capacity analysis, and risk discussions. They do not determine the actions a lender should take.
How Performance Is Evaluated
Performance is evaluated through rolling out-of-sample testing designed to reflect the product's monthly operating process. At each forecast step, the model uses information that would have been available as of that point and is evaluated against the subsequently realized outcome. The process is repeated across six evaluation windows, reducing dependence on the conditions of any single period. Reported results therefore measure recurring monthly forecast performance rather than assuming that one forecast remains unchanged for an entire year.
This paper uses the following measures:
- Mean absolute percentage error, or MAPE: The average absolute percentage difference between forecast and actual values A lower percentage reflects a more accurate forecast. For example, a 2.66% MAPE means that the average absolute forecast error was 2.66% of the realized value in the evaluation sample.
- R², or coefficient of determination: Measures how closely variation in forecast levels aligns with variation in realized outcomes within the evaluation sample. Higher values indicate greater explanatory fit. A result at or below zero indicates that the method did not improve upon the reference implied by the calculation.
Performance Compared with Historical Benchmarks
OBMMI Mortgage-Rate Forecast (as of July 2026)
| Method | MAPE | R² |
|---|---|---|
| Virtual Economist | 2.66% | 0.41 |
| 12-Month Trend Extrapolation | 5.36% | Below 0 |
| Mean Reversion | 6.06% | Below 0 |
| Persistence | 6.17% | Below 0 |
Across the evaluation windows, the Virtual Economist OBMMI forecast produced a 2.66% MAPE – approximately 50% lower average percentage error than the best-performing historical benchmark shown, which recorded a 5.36% MAPE. Virtual Economist also produced a positive Level R², while each displayed historical benchmark produced a result below zero.
The comparison illustrates the central advantage of the approach. Trend, mean-reversion, and persistence methods derive their forecasts primarily from the target's prior values. Virtual Economist can also evaluate forward-looking baseline inputs assembled from market-implied expectations, recognized public and industry forecasts, and documented assumptions.
Lock Volume Forecast
Lock Volume Forecast (as of July 2026)
The Lock Volume model translates the baseline OBMMI rate path into an expected monthly volume path using historical Optimal Blue lock activity and modeled time-series patterns. Because the rate forecast is an explicit model input, the volume outlook can respond when the baseline rate path changes rather than merely extending the recent volume trend.
The validation results shown here indicate an average 6.5% MAPE and an overall Level R² of 0.31 for the selected operating methodology. These figures describe model performance within the stated evaluation design; they do not guarantee future results.
Independent Review, Monitoring, and Responsible Use
Independent review
The Virtual Economist’s forecasting models are in the process of being subjected to independent third-party review. The review provides an additional layer of scrutiny alongside Optimal Blue's internal model development, documentation, testing, and monitoring processes.
Ongoing monitoring
Model results are compared with subsequently realized outcomes on a recurring basis, supporting evaluation of forecast performance as new observations become available and market conditions change.
Responsible-use guardrails
Virtual Economist produces mortgage rate and lock volume forecasts and evaluates economic scenarios to inform business planning. Defined scope, approved response guidance, user permissions, and monitoring processes support appropriate use.
Virtual Economist does not make, recommend, or automate business decisions. It does not set or recommend interest rates, margins, spreads, product pricing, lock volume or production targets, staffing levels, or hedging strategies. Those decisions remain entirely within the user’s judgment and existing governance processes.
Methodology and Data Notes
- Reported performance reflects rolling out-of-sample evaluation across six windows under the selected monthly operating methodology.
- At each forecast step, reported metrics use information available as of the applicable forecast date.
- Benchmark comparisons are limited to the methods and results displayed in this paper and should not be interpreted as a comparison with every external economic or mortgage forecast.
- Forecasts are conditional on their underlying baseline inputs and assumptions; actual outcomes may differ.
- Performance statistics and source assumptions should be accompanied by an “as of” date in the externally published version.
Virtual Economist is intended solely as an informational and analytical resource. Forecasts and scenarios are subject to uncertainty and should be considered together with other information, professional judgment, and a lender's established decision-making processes.