# Mortgage Market & Housing Risk from AI Displacement ## Research Summary **Source document**: CitriniResearch, "The 2028 Global Intelligence Crisis" (Feb 2026) **Core thesis**: AI-driven white-collar income displacement could threaten the $13T prime mortgage market -- not through bad origination (like 2008) but through structural income impairment *after* origination. **Date**: 2026-02-23 --- ## 1. The $13T Mortgage Market: Current State ### Market Size (Confirmed) The paper's $13T figure is accurate. Americans owe **$13.07 trillion** on 86.67 million mortgages, with balances growing by $98B to $13.2T in Q4 2025 (LendingTree, NY Fed Household Debt and Credit Report Q4 2025). ### Current Delinquency Rates Overall mortgage performance remains strong by historical standards, but cracks are emerging: - **Overall delinquency**: 4.26% (seasonally adjusted) as of Q4 2025, up 27bp QoQ and 28bp YoY (MBA) - **Conventional (prime) loans**: 2.89%, up 27bp QoQ -- still near record lows but trending up - **Single-family delinquency (commercial banks)**: 1.77% as of Q4 2024 (FRED) - **FHA loans**: Delinquency surged from 9.5% to 12% by Q2 2025, the highest since the pandemic; serious delinquencies rose from 3% to 4.8% - **Geographic divergence**: Lowest-income ZIP codes saw 90+ day delinquency rates surge sixfold from ~0.5% to ~3.0% between 2021 and late 2025, while wealthiest neighborhoods stayed flat (NY Fed Liberty Street Economics, Feb 2026) **Assessment**: The paper's claim about a fundamentally sound mortgage market built on income stability assumptions is correct. The $13T figure checks out. However, current stress is concentrated in lower-income and FHA segments, *not* in the prime segment the paper targets. Prime delinquencies remain historically low -- the crisis the paper describes has not yet materialized. --- ## 2. The Novel Mechanism: Post-Origination Income Impairment ### Claim: Loans Were Good at Origination This is the paper's most distinctive argument. Unlike 2008, where fraud and loose underwriting created bad loans from day one, this thesis posits that well-underwritten prime loans (780+ FICO, 20% down, verified income) become impaired when the borrower's income is structurally reduced by AI displacement. **Current underwriting standards support this framing:** - Fannie Mae requires income to be "stable and likely to continue for at least three years" (Fannie Mae Selling Guide B3-6-02) - Maximum DTI of 36%, extendable to 45% with compensating factors (credit score, reserves) - Employment stability is a key compensating factor for higher-DTI approvals - Income verification focuses on *current* earnings and recent history, not forward-looking industry vulnerability **Critical vulnerability**: Mortgage underwriting models fundamentally assume income continuity. They verify current employment and income trajectory but do not stress-test for sector-wide structural displacement. A borrower earning $250K/year in software engineering with a 780 FICO and 20% down will receive prime terms -- the model has no mechanism to discount for the possibility that AI agents may eliminate or substantially reduce demand for that role within 5-10 years. ### Academic Support: The "Double Trigger" Hypothesis Academic research on mortgage default (Elul et al., NBER; Dallas Fed; UC Chicago) strongly supports the mechanism the paper describes: - Mortgage default requires a "double trigger": negative equity AND income loss (typically unemployment) - **Structural unemployment** is a far more important predictor of default than cyclical unemployment - Mortgages are more sensitive to equity changes when structural unemployment is high - Critically, aggregate unemployment rates dramatically underestimate default risk compared to actual borrower-level job loss -- meaning current models may be systematically blind to this risk **Assessment**: The post-origination income impairment mechanism is theoretically sound and has academic support. The key question is whether AI displacement will actually produce the scale and concentration of income loss the paper projects. --- ## 3. Geographic Concentration Risk ### Tech/Finance Employment Concentration (Partially Confirmed) The paper claims >40% tech/finance employment in certain ZIP codes. Available metro-level data: - **San Francisco**: 22.54% of workforce in tech -- the highest among major cities - **San Jose**: 21.86% in tech - **Seattle**: 16.48% in tech (170,030 tech workers, 83.46 per 1,000 employees) - **Austin**: ~9% in tech (79,740 tech workers, 67.75 per 1,000) - **28% of all US tech jobs** are concentrated in just five cities: SF, Seattle, San Jose, LA, and Austin (Brookings) At the ZIP code level, the >40% figure is plausible for specific neighborhoods in SF (SOMA, Mission Bay), Seattle (South Lake Union, Capitol Hill), and San Jose (parts of North San Jose/Alviso near major campuses). These are exactly the areas with the highest mortgage values. ### Current Housing Market Conditions in Tech Hubs **San Francisco:** - Median home price: ~$1.29M (early 2025), up 1.8% YoY - Single-family homes up 12.5% YoY; condos down 8% - Inventory extremely tight: 1.6 months unsold inventory (Dec 2025) - **80% of SF homes lost value in recent period** (Iris Reads analysis) - 2022-2023 saw a 13.3% price decline ($1.46M to $1.27M) during tech layoffs - AI boom (OpenAI, Anthropic, Nvidia) has partially offset layoff effects **Seattle:** - Moderate price growth of 3-5% annually, slower than prior years - Inventory at 2.6 months, mortgage rates ~6.17% - Geographically constrained supply (water, mountains, zoning) - Home values declined 9% between May and August (specific year unclear from source) **Austin:** - Median sales price $435,000 in Dec 2025, **down 3.3% YoY** - The only major tech hub showing consistent price declines - Higher inventory than SF/Seattle; more responsive supply - Fundamental strengths (population growth, lifestyle) providing some floor ### The Paper's Price Decline Claims (11% SF, 9% Seattle, 8% Austin) These are the paper's *projected* declines in its 2028 scenario. As of early 2026: - SF experienced a 13.3% decline in 2022-2023 (proving such declines are possible) but has since partially recovered - Austin is currently declining at 3.3% YoY -- on a trajectory that *could* reach 8%+ if it accelerates - Seattle saw a brief 9% decline in mid-2025 between specific months **Assessment**: The geographic concentration is real. The price decline figures are not yet evident at the scale the paper projects, but precedent exists (SF 2022-2023). Austin is the weakest market and closest to validating the thesis. The extreme supply constraints in SF and Seattle (1.6 and 2.6 months respectively) provide a significant buffer against the kind of declines the paper projects. --- ## 4. "Invisible Stress" Indicators ### Credit Card Debt (Record Levels) - Total US credit card debt: **$1.277 trillion** (Q4 2025) -- highest since tracking began in 1999 - $350B higher than pre-pandemic record (Q4 2019) - Average balance per consumer: $6,523, up 2.2% YoY - Average APR: 22.25% (2025) - 38% of consumers say it's "difficult" or "very difficult" to pay bills on time - 46% of cardholders carry debt month-to-month ### HELOC Usage - HELOCs are the second most common use of home equity (after renovations) -- specifically for debt consolidation - Best HELOC rates under 7% vs. 22%+ credit card rates driving consolidation - This is consistent with the paper's claim about HELOC draws as a stress indicator ### Household Financial Stress - 43% of Americans cite money as negatively impacting mental health - Among those falling behind, 67% cite insufficient income - The economic divide is sharpening: stress concentrated in younger borrowers, lower-income households, and high-cost regions ### The "K-Shaped" Dynamic The NY Fed and CNBC report a "K-shaped" economic divide: wealthiest neighborhoods show flat delinquencies while lowest-income ZIP codes see surging defaults. This is *exactly the opposite* of what the paper predicts -- the paper argues it's the high-income prime borrowers who will be hit. However, the paper's thesis is that this K-shape will *invert* as AI displacement moves up the income ladder. **Assessment**: Household financial stress indicators are elevated and consistent with the paper's "invisible stress" narrative at the *macro* level. However, the stress is currently concentrated in lower-income segments, not in the prime/high-income segment the paper targets. The paper's thesis requires a *future* transmission of this stress upward into prime borrowers. Evidence for that transmission is not yet visible in mortgage data, though broader labor market data (see Section 5) provides some early signals. --- ## 5. AI Displacement: Scale and Timeline ### Current State of AI Job Displacement - **141,000+ tech positions** eliminated in 2025, with 31,000+ directly tied to AI restructuring - **1 million+ total layoffs** announced across all US industries in 2025 - Job cuts surged 175% in October 2025 vs. prior year - Share of workers using AI rose from 20% to 40% in two years - Job creation near zero over the past year (as of Feb 2026, per Fed Gov. Barr) ### Forecasts and Expert Views **Pessimistic:** - Microsoft analysis: 5 million white-collar jobs facing displacement - Goldman Sachs: 6-7% of US jobs could be displaced - Sen. warns unemployment could hit 25% among recent graduates - Fed Gov. Barr (Feb 17, 2026): Outlined a scenario where AI creates "a large share of the population [that] is essentially unemployable" - WEF: 92 million jobs displaced by 2030 **Optimistic:** - WEF: 170 million new jobs created by 2030 (net gain of 78 million) - AI delivers 66% average productivity increase across business tasks - Workers with AI skills command 56% wage premiums - Wages rising 2x faster in AI-exposed industries - Gartner: AI impact on global jobs neutral through 2026 - Goldman Sachs (gradual scenario): AI adds 0.3-0.9% to worker productivity; companies mostly reallocating and retraining rather than mass layoffs ### Fed's Three Scenarios (Barr, Feb 2026) 1. **Gradual adoption**: Modest productivity gains, limited displacement, manageable transition 2. **Rapid growth**: Massive productivity boom, significant displacement but eventually absorbed, upward pressure on neutral rate 3. **Doomsday**: Widespread unemployment, "essentially unemployable" population, financial instability Barr noted critically that the Fed can only address *cyclical* unemployment -- *structural* displacement from AI is outside the Fed's toolkit. **Assessment**: The displacement is real and accelerating, but the scale and speed remain deeply uncertain. The paper's 2028 timeline is aggressive. Current evidence supports a trajectory of gradual-to-moderate displacement that could accelerate. The key signal to watch is whether displacement moves beyond entry-level and lower-skilled roles into the $150K-$400K mid-career professionals who hold the largest prime mortgages. --- ## 6. The Savings Buffer and Delayed Delinquency ### The Paper's Claim High-income borrowers can maintain mortgage payments for quarters or even years after income loss by drawing on savings, 401(k)s, HELOCs, and credit cards -- creating a delay between job loss and visible mortgage stress. ### Supporting Evidence - Personal savings rate has been declining - HELOC-to-credit-card debt consolidation is at elevated levels - 401(k) hardship withdrawals have been trending up - The academic "double trigger" research confirms that negative equity *plus* income loss is required -- savings buffers extend the time before the trigger is pulled - High-income households have substantial buffers: median 401(k) balance for 55-64 year olds is ~$200K+; median home equity in tech hubs exceeds $500K ### Timeline Implications A household earning $300K/year with $200K in savings, $500K in home equity, and $150K in 401(k) could maintain a $3,000/month mortgage payment for 5+ years after total income loss, and indefinitely with partial income replacement. This makes the "invisible stress" period potentially very long. **Assessment**: This claim is highly plausible. The savings buffer effect would genuinely delay delinquency signals and is consistent with academic research. However, it also means the crisis would unfold more slowly than the paper's 2028 timeline suggests, unless displacement is both rapid and concentrated. --- ## 7. Counter-Arguments: Why the Thesis Could Be Wrong ### 1. Supply Constraints Provide a Price Floor - SF: 1.6 months inventory, well below the 6-month balanced market benchmark - Seattle: Geographically constrained (water, mountains), 2.6 months inventory - Even Austin, the weakest market, has fundamental demand drivers (population growth) - The national housing shortage is estimated at 3-7 million units - Builders are not overbuilding (unlike 2005-2007) **Implication**: Even significant demand reduction from income displacement may not produce the price declines the paper projects, because supply is so constrained. Prices are sticky downward when inventory is this low. ### 2. Remote Work Diversifies Geographic Risk - 20% of remote workers plan to relocate in 2025 - 53% of movers chose suburban areas - High-skilled workers have been redistributing from dense urban cores to suburbs and smaller cities - This *reduces* the geographic concentration the paper relies on -- tech workers are no longer all living in the same ZIP codes ### 3. The Mortgage Lock-In Effect - 24.6% of mortgage holders have rates below 3% (though declining) - 54% of homeowners wouldn't sell at any rate in 2025 - The lock-in effect suppresses selling, which suppresses inventory, which supports prices - However, this effect is fading: by early 2026, more holders are above 6% than below 3% - Financial pressure or job loss could *force* selling despite the rate penalty ### 4. PMI and Structural Protections - Loans with LTV >80% require PMI, which cuts loss severity by 22 percentage points - The paper's scenario specifically targets borrowers with 20% down (no PMI) -- these borrowers have substantial equity buffers - GSE (Fannie/Freddie) capital requirements have been strengthened since 2008 - Mortgage insurance industry is better capitalized than pre-2008 ### 5. AI Creates Jobs and Raises Productivity - WEF projects net gain of 78 million jobs by 2030 - AI-exposed industries see wages rising 2x faster - Workers acquiring AI skills command 56% wage premiums - Many displaced workers may transition to AI-augmented or AI-adjacent roles - Historical precedent: prior technology waves (ATMs, spreadsheets, internet) displaced specific roles but expanded overall employment ### 6. Regulatory and Policy Responses - The Fed is actively monitoring AI labor market risks (Barr's Feb 2026 speech) - Mortgage forbearance programs (proven during COVID) could be deployed - Congress has tools for housing market intervention - Fannie/Freddie could adjust underwriting standards prospectively ### 7. The Displacement Timeline May Be Slower Than Feared - Goldman Sachs' base case: gradual adoption, modest productivity gains - Gartner: AI impact on global jobs neutral through 2026 - Enterprise AI adoption faces implementation friction, organizational resistance, regulatory constraints - The paper's 2028 crisis assumes very rapid, concentrated displacement -- history suggests technology adoption curves are typically longer --- ## 8. Early Warning Indicators to Watch Investors should monitor the following signals for early evidence that the paper's thesis is materializing: ### Tier 1: Direct Mortgage Stress Signals 1. **Prime mortgage delinquency rates by metro**: Watch for any uptick in 30-day delinquencies specifically in SF, Seattle, Austin, NYC -- currently at historic lows 2. **Mortgage delinquency by borrower income quartile**: The NY Fed's geographic analysis should be supplemented with income-stratified data; watch for the "K-shape" to begin inverting 3. **Foreclosure filings in tech-heavy ZIP codes**: Currently minimal; any increase is a red flag 4. **Jumbo mortgage performance**: Jumbo loans are concentrated in tech hubs; separate tracking from conforming loans ### Tier 2: Precursor Stress Signals ("Invisible Stress") 5. **HELOC origination volumes in tech metros**: Spike in HELOC draws by prime borrowers could signal income replacement needs 6. **401(k) hardship withdrawal rates**: Monitor by employer sector (tech, finance, professional services) 7. **Credit card utilization rates among high-FICO borrowers**: Rising utilization among 750+ FICO holders would be a canary 8. **Debt-to-income ratio changes on existing mortgages**: Not directly observed, but credit bureau data on total debt growth among mortgage holders is a proxy ### Tier 3: Labor Market Signals 9. **Tech sector unemployment rate**: Currently low but rising; watch for acceleration 10. **White-collar job postings vs. AI tool adoption**: Declining postings in roles where AI tools are being deployed 11. **Income replacement rates after tech layoffs**: Are displaced workers finding comparable-income roles? The 2022-2023 layoffs showed most did; watch if this changes 12. **Contract/gig economy growth among former salaried professionals**: Indicator of the "downshift" the paper describes ### Tier 4: Market Signals 13. **Mortgage-backed securities spreads**: Widening spreads on RMBS tranches with tech-metro concentration 14. **Home price trends in tech-heavy metros vs. national average**: Divergence would validate geographic concentration risk 15. **Mortgage REIT performance**: These entities are directly exposed to the risk --- ## 9. Plausibility Assessment ### What the Paper Gets Right - **The fundamental vulnerability is real**: Mortgage underwriting assumes income continuity; AI displacement could break that assumption - **The mechanism is novel and underappreciated**: Unlike 2008, this isn't about bad origination -- it's about the world changing after loans are written - **Geographic concentration exists**: Tech/finance employment is genuinely concentrated in specific metros and ZIP codes - **The savings buffer would delay signals**: High-income borrowers would mask stress for quarters or years - **The resolution path is genuinely unclear**: You can't regulate AI away without sacrificing productivity gains - **Fed officials are taking the risk seriously**: Gov. Barr's Feb 2026 speech explicitly outlined the "essentially unemployable" scenario ### What the Paper Gets Wrong or Overstates - **The timeline is too aggressive**: A full-blown crisis by 2028 requires displacement at a pace not yet supported by evidence. More realistic: early stress signals by 2027-2028, potential crisis by 2029-2031 - **Supply constraints are underweighted**: The paper doesn't adequately address how tight inventory (1.6 months in SF) would resist price declines - **Remote work diversification is ignored**: Geographic concentration is declining, not increasing - **Job creation and retraining are dismissed**: Historical precedent and current data show substantial job creation in AI-adjacent fields - **The 11% SF price decline claim**: SF already experienced a 13.3% decline in 2022-2023 and recovered; supply constraints make a sustained decline of this magnitude unlikely without massive forced selling - **Not all white-collar jobs are equally exposed**: Many professional roles (healthcare, legal, education) have regulatory and institutional barriers to rapid AI displacement ### Overall Plausibility Rating: Moderate-to-High Conceptual, Low-to-Moderate on Timeline The *mechanism* the paper describes is sound and represents a genuine, underappreciated risk to the mortgage market. The academic literature on structural unemployment and mortgage default supports the theoretical framework. Fed officials are taking the risk seriously. However, the *timeline* (crisis by mid-2028) requires a pace of displacement that exceeds current evidence. The paper's price decline projections underweight supply constraints and the mortgage lock-in effect. Current data shows mortgage stress concentrated in lower-income segments, not the prime segment the paper targets. **Most likely scenario**: AI displacement proceeds at a pace between "gradual" and "rapid," producing visible stress in prime mortgage markets in specific tech-heavy metros by 2028-2030, but not the systemic crisis the paper describes until the early 2030s (if at all). The risk is asymmetric -- if displacement accelerates faster than expected, the consequences would be severe precisely because the mortgage market is not pricing this risk. --- ## Key Data Sources - MBA National Delinquency Survey, Q4 2025 - NY Fed Household Debt and Credit Report, Q4 2025 - NY Fed Liberty Street Economics, "Where Are Mortgage Delinquencies Rising the Most?" (Feb 2026) - Fannie Mae Selling Guide B3-6-02 (DTI ratios) - Chicago Fed, "Tail Risk for Banks Posed by Investments in Generative AI" (Feb 2026) - Fed Gov. Barr, "AI and the Labor Market" speech (Feb 17, 2026) - LendingTree, Bankrate, NerdWallet -- consumer debt statistics - Redfin, Realtor.com, Norada Real Estate -- housing market data - Goldman Sachs Research -- AI labor market impact - World Economic Forum -- Future of Jobs Report - Brookings Institution -- tech job geographic concentration - Course Report -- tech worker geographic distribution - Urban Institute -- PMI loss severity analysis --- *Disclaimer: This is educational research and analysis, not investment advice. Markets involve risk. Past performance does not guarantee future results.*