---
name: Delinquency Risk Analysis Expert
description: Delinquency risk, at-risk tenants, collection risk, bad debt, payment risk, tenant risk score, collections strategy, eviction cost, payment plan, FDCPA, TCPA, risk scoring model, early intervention.
---

# Delinquency Risk Analysis Expert

## Table of Contents
1. [Sunrise Risk Scoring Model](#sunrise-risk-scoring-model)
2. [Feature Engineering](#feature-engineering)
3. [Risk Tier Framework](#risk-tier-framework)
4. [Collection Strategy Matrix](#collection-strategy-matrix)
5. [Early Warning System](#early-warning-system)
6. [Legal Requirements for Collections](#legal-requirements-for-collections)
7. [Payment Plan Design](#payment-plan-design)
8. [Eviction Cost-Benefit Analysis](#eviction-cost-benefit-analysis)
9. [Seasonal Adjustment Methodology](#seasonal-adjustment-methodology)
10. [autoMHatic Referral Integration](#automhatic-referral-integration)
11. [Bad Debt Reserve Calculation](#bad-debt-reserve-calculation)
12. [Model Performance Monitoring](#model-performance-monitoring)
13. [Ethical AI & Fair Housing Guardrails](#ethical-ai--fair-housing-guardrails)
14. [Decision Trees](#decision-trees)
15. [Common Gotchas](#common-gotchas)
16. [Output Formats](#output-formats)
17. [References](#references)

---

> **Workflow handoffs:** This skill participates in the **Eviction workflow:** `mhp-regulatory` (notice requirements) → `delinquency-model` (risk scoring, collection escalation) → `resident-communications` (notices/letters) → `fair-housing` (compliance check)

## Sunrise Risk Scoring Model

### Purpose
Score and analyze tenant delinquency risk to enable proactive collection interventions and reduce bad debt. The model runs monthly via `delinquency_risk_model.py`, outputs to Google Sheets, and targets a portfolio-wide delinquency rate below 3% with a collection rate of 98%+.

### Architecture Overview
```
RentManager API → CSV Exports (reports/YYYY-MM/) → Risk Model (Python)
    ↓                                                     ↓
BLS API (unemployment) ──────────────────────────→ Risk Scores
    ↓                                                     ↓
Probability Engine (clusters, economic state) ──→ Google Sheets Dashboard
```

**Source code:** `/Operations/delinquency-risk-model/delinquency_risk_model.py`
**Output sheet:** Google Sheet ID `1kFhhIqICSg-fpUYMdYjgFLm9p7PMa97SbCIy8WHDxUM`
**Tabs:** `Risk_Scores`, `Action_List`, `Summary`

### Score Components (v2.0 — 0-100+ scale)

The v2.0 model uses six scoring components plus an economic state modifier:

| Factor | Max Points | Weight | Data Source |
|--------|-----------|--------|-------------|
| **Payment History** | 35 | Core predictor | RentManager Payments (12-month lookback) |
| **Balance Aging** | 25 | Current status | RentManager DelinquencyAging report |
| **Economic Conditions** | 15 | External risk | BLS API (state unemployment rates) |
| **Property Cluster** | 10 | Portfolio context | Probability Engine Markov Chain model |
| **Tenure** | 10 | Stability proxy | RentManager Leases (MoveInDate) |
| **Lease Expiration** | 5 | Flight risk | RentManager Leases (ExpectedMoveOutDate) |
| **Economic State Modifier** | +10 | Regime overlay | SLOWDOWN: +5, RECESSION: +10, TROUGH: +8, RECOVERY: +3 |

**Note on skill vs. code:** The original skill documented weights as Payment 40%, Balance 25%, Tenure 15%, Economic 10%, Seasonal 10%. The actual v2.0 code uses the component breakdown above. When discussing the model, use the v2.0 code values. The skill weights are approximate conceptual groupings.

### Scoring Formula (from code)
```python
base_score = (
    int(payment_score * 0.875)   # Scale 40→35 max
    + balance_score               # 25 max (unchanged)
    + int(tenure_score * 0.67)    # Scale 15→10 max
    + int(economic_score * 0.75)  # Scale 20→15 max
    + cluster_score               # 10 max (new in v2.0)
    + lease_score                 # 5 max (new in v2.0)
)
total_score = base_score + state_modifier  # Can exceed 100 in RECESSION
```

### Home Ownership Classification (v2.0)

The model classifies tenants by home ownership, which affects flight risk:

| Type | Description | Risk Implication |
|------|-------------|------------------|
| **TOH** (Tenant-Owned Home) | Resident owns mobile home, pays lot rent | Lower flight risk — significant sunk cost to relocate |
| **COH** (Community-Owned Home) | Community owns home, resident rents | Higher flight risk — can leave with minimal friction |
| **OTHER** | Storage, garage, RV, etc. | Excluded from risk model |

TOH UnitTypeIDs: 75, 96, 98, 99, 100, 105, 107
COH UnitTypeIDs: 77, 78, 79, 80, 84, 87, 103, 104, 117

### Property Cluster Assignments

From the Probability Engine's Markov Chain model. Clusters describe how property occupancy responds to economic stress:

| Cluster | Properties | Risk Effect |
|---------|-----------|-------------|
| **RESILIENT** | Darby, Ridgebrook Hills, STM, Walston, Bellecrest, MNL, Lakeridge | 0 pts — occupancy UP when economy DOWN |
| **NEUTRAL** | Elk, Tilghman, APE, Cedarhurst | 5 pts — uncorrelated |
| **SENSITIVE** | Sherman, Mancuso, Rolling Hills, Timberview, Dutch Gardens, Miami, Park Estates | 10 pts — occupancy DOWN when economy DOWN |

During SLOWDOWN/RECESSION/TROUGH, cluster scoring amplifies: SENSITIVE properties get max 10 pts, RESILIENT get 0 pts regardless of base.

### Economic State Model

Five-state regime model integrated from the Probability Engine:

| State | Modifier | Trigger |
|-------|----------|---------|
| EXPANSION | +0 | Avg unemployment < 4.5% |
| RECOVERY | +3 | Avg unemployment 4.5-5.0% |
| SLOWDOWN | +5 | Avg unemployment 5.0-6.0% |
| TROUGH | +8 | Peak unemployment, stabilizing |
| RECESSION | +10 | Avg unemployment > 6.0% |

---

## Feature Engineering

### Payment History (35 pts max) — Why It Is the Strongest Predictor

Payment history carries the most weight because past behavior is the single best predictor of future behavior. Research on delinquency prediction consistently finds that recency, frequency, and severity of late payments dominate other features. The model uses a 12-month lookback with day-of-month analysis.

**Scoring logic (from code):**

| Pattern | Raw Score (0-40) | Scaled (×0.875) |
|---------|-----------------|------------------|
| Perfect payment history | 0 | 0 |
| 1-2 late payments (paid after 5th) | 10 | 8 |
| 3-4 late payments | 20 | 17 |
| 5-6 late payments | 30 | 26 |
| 7+ late payments | 40 | 35 |
| Worsening trend (>50% late in last 90 days) | +5 bonus | +4 |

**Feature engineering notes:**
- "Late" is defined as payment received after the 5th of the month (grace period assumption)
- No payment history at all = 20 raw pts (medium risk) — avoids penalizing data gaps
- No recent payments (within lookback) = 25 raw pts — slightly higher than no history
- Trend detection: if >50% of last 90 days' payments are late, add 5 pts (capped at 40)
- Future enhancement: weight by recency (exponential decay), distinguish partial vs. full payments

**Evidence base:** CFPB research (Jan 2025) found that 23% of renters incurred late fees at peak, declining to 14% by late 2024. Late fee incidence is the dominant driver of delinquency classification. Seasonal patterns in late payment rates (spring tax refund dip, holiday stress spike) are well-documented but shifting post-COVID.

### Balance Aging (25 pts max) — Current Financial Distress

Measures how deep the tenant is currently in arrears. Uses RentManager's DelinquencyAging buckets.

| Aging Bucket | Points | Rationale |
|-------------|--------|-----------|
| Current (no delinquency) | 0 | No current distress signal |
| 0-30 days past due | 5 | Common temporary cash flow issue |
| 31-60 days | 15 | Entering serious delinquency |
| 61-90 days | 20 | High likelihood of escalation |
| 90+ days | 25 | Near-certain write-off candidate |

**Why this matters:** Balance aging is a lagging indicator — by the time a tenant hits 90+ days, intervention options narrow significantly. The payment history score acts as the leading indicator; balance aging confirms the crisis.

### Tenure (10 pts max) — Stability Proxy

Shorter tenancy correlates with higher delinquency risk across all multifamily asset classes. In MHPs, TOH residents with 5+ years tenure have extremely low default rates because of the cost and difficulty of relocating a manufactured home.

| Tenure | Raw Score (0-15) | Scaled (×0.67) |
|--------|-----------------|------------------|
| 3+ years | 0 | 0 |
| 1-3 years | 5 | 3 |
| 6 months - 1 year | 10 | 6 |
| < 6 months | 15 | 10 |

**MHP-specific insight:** Unlike apartments (12-16 month average stay), MHP residents average 10-14 years. A new resident (<6 months) in an MHP is statistically much more unusual and carries proportionally higher risk than in conventional multifamily.

### Economic Conditions (15 pts max) — External Stress

Uses BLS state-level unemployment rates via API, with a 30-day cache. Maps property zip codes to states.

| Unemployment Rate | Raw Score (0-20) | Scaled (×0.75) |
|-------------------|-----------------|------------------|
| < 4% | 0 | 0 |
| 4-5% | 5 | 3 |
| 5-6% | 10 | 7 |
| 6-8% | 15 | 11 |
| 8%+ | 20 | 15 |

**Property-to-state mapping:** Uses zip code prefix to determine state (e.g., "43xxx" → OH, "46xxx" → IN). Fallback rates hardcoded if API fails.

### Property Cluster (10 pts max) — Portfolio Risk Context

Derived from the Probability Engine's Markov Chain analysis of occupancy-vs-economy correlation. During economic stress periods, SENSITIVE properties see occupancy declines (increasing competitive pressure on remaining residents), while RESILIENT properties may actually see occupancy increases (affordable housing demand surge).

### Lease Expiration (5 pts max) — Flight Risk

| Days to Expiration | Points |
|--------------------|--------|
| Month-to-month / expired | 2 |
| 90+ days | 0 |
| 60-90 days | 3 |
| 30-60 days | 4 |
| < 30 days | 5 |

**MHP context:** Most MHP leases are month-to-month or annual. For TOH residents, lease expiration is less meaningful because they cannot easily leave. For COH residents, lease expiration is a genuine flight risk signal.

---

## Risk Tier Framework

### Tier Definitions (v2.0 Calibration)

Calibrated for approximately 50/35/12/3 distribution during EXPANSION. During economic stress, the state modifier naturally shifts more tenants into higher tiers.

| Tier | Score Range | Target Distribution | Description |
|------|-------------|-------------------|-------------|
| **LOW** | 0-30 | ~50% | Stable tenants, minor or no issues |
| **MEDIUM** | 31-45 | ~35% | Requires monitoring, proactive check-ins |
| **HIGH** | 46-55 | ~12% | Active outreach needed, payment plan discussions |
| **CRITICAL** | 56+ | ~3% | Immediate intervention, formal collections process |

### Action Protocols by Tier

#### CRITICAL (56+) — Immediate Intervention
- **Day 1:** Manager phone call within 24 hours
- **Day 1-3:** Document payment commitment or refusal
- **Day 3-5:** If no contact: formal demand letter (certified mail + posted on door)
- **Day 5-7:** Offer payment plan with written agreement
- **Day 7-14:** If no resolution: consult with legal, prepare notice to quit
- **Day 14+:** Initiate eviction filing (state-specific timeline)
- **Ongoing:** Weekly status update in tenant file, flag for property manager review
- **Minimum balance threshold for Action List:** $200 (below this, outreach cost exceeds recovery value)

#### HIGH (46-55) — Active Outreach
- **Week 1:** Personal phone call, text, or in-person visit
- **Week 1-2:** Review account for billing errors or disputes
- **Week 2:** Discuss payment options, document conversation
- **Week 2-4:** If improving: continue monitoring. If worsening: escalate to CRITICAL protocol.
- **Monthly:** Property manager review for 3 months minimum

#### MEDIUM (31-45) — Monitoring
- Monthly check-in call or text
- Standard late notices per property protocol
- Monitor for trend direction (improving vs. worsening)
- Quarterly review with property manager

#### LOW (0-30) — Standard Operations
- Automated payment reminders only
- Annual review
- No active intervention required

### Escalation Triggers (Tier Upgrades)
Move a tenant UP one tier immediately if:
- Balance increases by >50% month-over-month
- Tenant misses two consecutive payments after a payment plan agreement
- Legal notice returned undeliverable (skip trace needed)
- Tenant files bankruptcy or invokes SCRA protections (special handling required)
- Property receives code violation tied to the tenant's lot

---

## Collection Strategy Matrix

### By Risk Tier and Balance Amount

| Tier | Balance < $500 | Balance $500-$2,000 | Balance > $2,000 |
|------|---------------|---------------------|-------------------|
| **CRITICAL** | Demand letter + payment plan | Attorney letter + payment plan | Legal consultation, eviction evaluation |
| **HIGH** | Phone call + text reminder | Payment plan offer + written agreement | Manager meeting + formal payment plan |
| **MEDIUM** | Automated reminder | Phone call + payment options | Review for billing errors, then phone call |
| **LOW** | Standard late notice | Standard late notice | Verify accuracy, then standard notice |

### By Home Ownership Type

**TOH (Tenant-Owned Home):**
- Higher leverage — resident has invested in the home and lot improvements
- Eviction is MORE impactful (they must sell or move a manufactured home, costing $5,000-$15,000+)
- Payment plans are more appropriate because flight risk is lower
- Preferred approach: longer payment plans (3-6 months), preserve the tenancy
- autoMHatic referral: strong candidates for home improvement loans if they have good history minus a temporary hardship

**COH (Community-Owned Home):**
- Lower leverage — resident can leave with 30 days' notice in most states
- Eviction is LESS costly (no home to deal with, unit is turnover-ready faster)
- Shorter payment plan timelines (30-60 days max)
- Focus on rapid resolution or rapid turnover
- Higher priority for in-person visits (flight risk is real)

### By Property Cluster

**SENSITIVE properties** (Sherman, Mancuso, Rolling Hills, Timberview, Dutch Gardens, Miami, Park Estates):
- During economic stress: be MORE flexible with payment plans (these communities are harder to backfill)
- Vacancy at a SENSITIVE property during a downturn is doubly costly
- Consider rent concessions for long-term TOH residents

**RESILIENT properties** (Darby, Ridgebrook Hills, STM, Walston, Bellecrest, MNL, Lakeridge):
- Standard collections approach — backfill is easier
- Less need for concessions; market demand supports the lot rent

### Communication Channel Effectiveness
```
Most effective (by response rate, industry research):
1. In-person visit at the home — 65-80% response
2. Phone call (morning, 9-11am) — 40-55% response
3. Text message — 35-50% response
4. Door notice — 30-40% response
5. Email — 15-25% response
6. Mailed letter — 10-20% response (but creates legal paper trail)
```

**Best practice:** Lead with phone/text for MEDIUM and HIGH. Use certified mail + door posting for CRITICAL (creates the legal paper trail needed for eviction). Never rely on email alone for serious delinquency — many MHP residents do not check email regularly.

---

## Early Warning System

### Leading Indicators (Detect Before Delinquency Appears)

| Indicator | Detection Method | Risk Signal |
|-----------|-----------------|-------------|
| **Partial payments** | Payment amount < monthly charge for 2+ months | Tenant stretching to pay — likely to fall behind |
| **Payment date drift** | Average payment date moving later each month | Cash flow tightening |
| **NSF/bounced payments** | RentManager transaction reversals | Bank account distress |
| **Maintenance request spike** | Unusually high requests from one lot | Possible preparation for dispute or move-out |
| **Utility payment issues** | If community-billed: utility balance growing | Prioritizing other expenses over housing |
| **Lease non-renewal** | Tenant declines renewal or does not respond | Planning to leave |
| **New residents at same lot** | Unauthorized occupants detected | Income instability, subletting risk |
| **Communication avoidance** | Tenant not answering calls/texts after previously responsive | Classic pre-default behavior |

### Trend Detection Algorithms

**Payment Date Drift:** Track the 3-month rolling average of payment day-of-month. If the average moves from day 3 to day 10 over 3 months, flag as "drifting." If it exceeds day 15, flag as "high drift risk."

**Partial Payment Pattern:** If a tenant pays less than 90% of monthly charges for 2 consecutive months, flag as "partial payment pattern." This often precedes full non-payment by 30-60 days.

**Worsening Trend Bonus (in code):** The model already adds 5 points if >50% of the last 90 days' payments are late. This catches tenants whose behavior is deteriorating even if their 12-month average looks acceptable.

### Alert Triggers for Property Managers

Generate alerts when:
- Any tenant moves from LOW/MEDIUM to HIGH or CRITICAL
- A CRITICAL tenant's balance exceeds 2x monthly rent
- A HIGH tenant has been HIGH for 3+ consecutive months without improvement
- Any property has >5% of tenants in HIGH/CRITICAL combined
- Portfolio-wide delinquency rate approaches 3% target threshold
- A tenant on a payment plan misses a scheduled installment

---

## Legal Requirements for Collections

*Last verified: March 2026. Review trigger: any legislative session in a Sunrise state, or quarterly at minimum.*

### State Eviction Notice Periods (Nonpayment of Rent)

| State | Notice Period | Cure Right | Special MH Rules | Est. Total Timeline |
|-------|--------------|------------|-------------------|---------------------|
| **OH** | 3 days | No automatic cure | ORC 1923 applies to MH lots | 30-45 days |
| **IN** | 10 days | Yes, can pay to cure | IC 32-31-3 | 45-60 days |
| **MD** | 10 days | Yes, can pay to cure | Real Property §8-402 | 30-60 days |
| **MI** | 7 days (demand for possession) | No | MCL 600.5714 | 30-45 days |
| **FL** | 3 days | Yes, can pay within 3 days | F.S. §83.56; 90-day notice for rent increase | 30-45 days |
| **WV** | 5 days | Yes | WV Code §37-6-5 | 30-60 days |
| **PA** | 10 days | No automatic cure | 68 P.S. §250.501 | 30-60 days |
| **AL** | 7 business days | Yes, can pay to cure | Code §35-9A-421 | 30-45 days |
| **GA** | None required (demand can be immediate) | No | O.C.G.A. §44-7-50 | 28-45 days |
| **AR** | 3 days (if 5+ days past due) | No | A.C.A. §18-60-304 | 30-45 days |
| **MN** | 14 days (at-will) | No | M.S. §504B.135; 90-day rent increase notice | 30-60 days |
| **ND** | 3 days | No | NDCC §33-06-01 | 21-45 days |
| **WI** | 5 days (lease < 1 yr), 30 days (lease > 1 yr) | Yes (5-day notice) | Wis. Stat. §704.17 | 30-60 days |
| **MO** | 5 days | No | RSMo §441.060 | 21-30 days |
| **IL** | 5 days | Yes, can pay within 5 days | 735 ILCS 5/9-209 | 30-60 days |

**CRITICAL:** Many states have ADDITIONAL manufactured housing-specific protections (longer notice periods, relocation assistance requirements, right of first refusal on home sale). Always check state MH statutes separately from general landlord-tenant law.

### FDCPA Compliance (Fair Debt Collection Practices Act)

*Last verified: March 2026.*

**Does FDCPA apply to Sunrise?** It depends on who is collecting:
- **Property managers collecting current rent:** Generally NOT covered (collecting own debt)
- **Property managers collecting for a DIFFERENT owner:** MAY be covered as a third-party debt collector
- **Collection agencies or attorneys collecting past-due rent:** COVERED
- **In-house staff collecting AFTER tenant moves out:** Courts are split; treat as covered to be safe

**Key FDCPA requirements if applicable:**
1. **Validation notice:** Within 5 days of initial contact, send written notice stating: amount owed, name of creditor, and tenant's right to dispute within 30 days
2. **Contact hours:** Only between 8:00 AM and 9:00 PM (tenant's local time)
3. **Cease contact:** If tenant requests in writing, all contact must stop (except to notify of legal action)
4. **No harassment:** No threats, abusive language, or false/misleading statements
5. **No third-party disclosure:** Cannot discuss the debt with anyone other than the tenant, their attorney, or a credit bureau
6. **Debt collector identification:** Must identify themselves as a debt collector in every communication

**Practical guidance for site managers:** Even when FDCPA technically does not apply, follow its principles. It costs nothing extra and creates defensible documentation. If a tenant claims FDCPA violation, the burden shifts to Sunrise to prove exemption.

### TCPA Compliance (Telephone Consumer Protection Act)

*Last verified: March 2026.*

**Text messages to tenants about rent:**
- Require **prior express consent** (get this at lease signing)
- Must include **opt-out mechanism** in every text ("Reply STOP to unsubscribe")
- Autodialed calls to cell phones require prior express consent
- **Safe harbor:** Include a TCPA consent clause in the lease agreement covering payment reminders, community notices, and emergency communications
- Violations: $500-$1,500 per unsolicited text/call — adds up fast in a portfolio

**Recommended lease clause (have legal counsel review):**
> "Resident consents to receive communications from Management, including payment reminders, community notices, and emergency alerts, via text message, email, and phone calls to the number(s) provided. Resident may revoke this consent at any time by providing written notice to the community office."

### Bankruptcy Protections

When a tenant files bankruptcy:
1. **Automatic stay** takes effect IMMEDIATELY — all collection activity must stop
2. No new eviction filings, no demand letters, no phone calls about the debt
3. Must file a **motion for relief from stay** in bankruptcy court to proceed with eviction
4. **Chapter 7:** Tenant may surrender the lease; past-due rent becomes unsecured claim
5. **Chapter 13:** Tenant may propose a repayment plan; must stay current on post-petition rent
6. **Practical impact:** Adds 30-90 days to any collection/eviction timeline
7. **Document everything:** Note bankruptcy filing date, case number, attorney name in tenant file

**Flag in risk model:** Any tenant with an active bankruptcy should be tagged separately — they are outside normal collection protocols.

### SCRA Protections (Servicemembers Civil Relief Act)

*Last verified: March 2026.*

When a tenant is an active-duty servicemember:
- **Cannot evict** without a court order for nonpayment of rent (if monthly rent ≤ inflation-adjusted threshold, currently ~$4,700/month)
- Court may **stay eviction for up to 90 days** or longer
- Court may order partial garnishment of military pay as equitable relief to landlord
- Servicemember can **terminate lease early** upon receiving deployment orders or PCS
- **Verification:** Use the DOD SCRA website (scra.dmdc.osd.mil) to check active-duty status before initiating any eviction

**Practical impact for MHPs:** Unlikely to apply often (most MHP lot rents are well below the threshold), but must check. A single SCRA violation carries federal penalties and reputational damage.

### Fair Housing in Collections

> **Cross-reference:** See `fair-housing` skill for comprehensive protected class definitions, disparate impact analysis, and state-by-state protections. See `resident-communications` skill Section 6 for collections letter templates and approved/prohibited language.

**Collection practices that create Fair Housing risk:**
- Inconsistent enforcement: pursuing delinquency from some residents but not others without documented business justification
- Language barriers: failing to provide notices in languages spoken by significant resident populations
- Disability accommodations: refusing to accept third-party payments, alternative payment schedules, or reasonable modifications to collection procedures
- Familial status: treating families with children differently in payment plan terms

**Mitigation:** Apply collection protocols uniformly. Document every interaction. Use the same escalation timeline for every tenant at the same tier. If an exception is made, document the business justification.

---

## Payment Plan Design

### Structuring Effective Payment Plans

**Core principles:**
1. **Affordability:** Total monthly payment (rent + arrears catch-up) should not exceed 110% of normal rent for most residents. If it does, the plan is likely to fail.
2. **Written agreement:** Always get the plan in writing with both parties' signatures. Include specific dates and amounts.
3. **Front-loaded good faith:** Require a down payment (typically 25-50% of past-due amount) as evidence of commitment.
4. **Short duration:** 30-90 days for COH, up to 6 months for TOH (longer for TOH because they are lower flight risk).
5. **Clear consequences:** State explicitly what happens if a payment is missed (revert to full balance due, resume eviction process).

### Payment Plan Templates by Scenario

**Scenario A — Temporary hardship (job loss, medical event):**
```
Down payment: 25% of past-due amount
Remaining balance: Split evenly over 3-4 months
Added to regular monthly rent
Condition: Must remain current on new charges
Failure clause: One missed installment = full balance due + eviction process resumes
```

**Scenario B — Chronic lateness (pattern of 10-15 day late payments):**
```
No down payment required (they eventually pay)
Switch to auto-pay or direct debit
Require payment by the 1st, late fee on the 6th (per lease)
3-month monitoring period
Condition: If 2+ late payments in monitoring period, escalate to HIGH tier
```

**Scenario C — Deep delinquency (60+ days, $2,000+):**
```
Down payment: 50% of past-due amount within 7 days
Remaining: 4-6 equal monthly installments
Written agreement reviewed by regional manager
Condition: Must remain current on ALL new charges
Failure clause: One missed installment = immediate legal consultation
COH tenants: Consider surrender/move-out agreement as alternative
```

### Monitoring Payment Plan Compliance

- Track payment plan installments as separate line items in RentManager (or notes)
- Set calendar reminders for each installment date
- If a tenant misses one installment: immediate phone call (same day)
- If a tenant misses two installments: plan is in default — revert to full collection/eviction path
- Monthly review: is the tenant's total balance decreasing? If not, the plan is not working

### When to Offer vs. When to Skip Payment Plans

**Offer a payment plan when:**
- First-time delinquency with documented hardship (job loss, medical, family emergency)
- TOH resident with 2+ years tenure (high investment in staying)
- Balance is less than 3 months' rent
- Tenant is communicative and willing to engage

**Skip the payment plan when:**
- Tenant has defaulted on a previous payment plan
- Tenant is non-communicative (no response to 3+ contact attempts)
- COH resident with <6 months tenure and growing balance
- Balance exceeds 3 months' rent with no down payment capability
- Evidence of lease violations beyond nonpayment (unauthorized occupants, property damage)

---

## Eviction Cost-Benefit Analysis

### True Cost of Eviction in MHPs

Eviction in a manufactured housing community is fundamentally different from apartment eviction. The analysis must consider the home's disposition.

**TOH Eviction Cost Breakdown:**

| Cost Component | Estimate | Notes |
|----------------|----------|-------|
| Legal fees (uncontested) | $500-$1,500 | Attorney filing + court appearance |
| Legal fees (contested) | $2,000-$10,000+ | Discovery, hearings, continuances, appeals |
| Court costs | $50-$500 | Filing fees, service of process |
| Lost rent during process | $1,500-$6,000 | 2-4 months at avg lot rent of $400-$600 |
| Post-eviction vacancy | $1,200-$3,600 | 2-6 months to find new TOH tenant or sell lot |
| Abandoned home disposition | $3,000-$15,000 | Title issues, demo costs, or home rehab |
| Staff time | $500-$1,500 | Manager hours for process, court, cleanup |
| **Total TOH eviction** | **$7,000-$35,000+** | |

**COH Eviction Cost Breakdown:**

| Cost Component | Estimate | Notes |
|----------------|----------|-------|
| Legal fees | $500-$5,000 | Same as above |
| Court costs | $50-$500 | Same as above |
| Lost rent during process | $2,000-$8,000 | 2-4 months (lot rent + home rent higher) |
| Unit turnover (rehab) | $2,000-$8,000 | Cleaning, repairs, appliances |
| Re-leasing vacancy | $1,000-$3,000 | 1-3 months marketing + showing |
| Staff time | $500-$1,500 | Same as above |
| **Total COH eviction** | **$6,000-$26,000** | |

### Break-Even Analysis: Eviction vs. Payment Plan

**Rule of thumb:** If the total owed is less than the estimated eviction cost, a payment plan is financially superior — even if only 60-70% of the arrears are ultimately collected.

```
Break-even formula:
  If (Past-Due Balance × Collection Probability) > Eviction Cost
    → Eviction may be justified
  Else
    → Payment plan or negotiated move-out is cheaper

Example:
  Tenant owes $3,000. Eviction cost estimate: $8,000 (TOH).
  Even if payment plan collects only 70% ($2,100):
    Payment plan cost: $0 legal + $900 write-off = $900 net loss
    Eviction cost: $8,000 + $3,000 write-off = $11,000 net loss
  → Payment plan saves $10,100
```

### When Eviction IS Financially Justified

- Balance exceeds 6+ months rent AND tenant is non-communicative
- Tenant has caused property damage that compounds monthly
- Tenant's behavior is affecting neighboring residents (noise, safety, code violations) — the cost is retention risk for other tenants
- COH unit where the home is in good condition and can be re-rented quickly (low vacancy cost)
- Market conditions: strong demand for lots at this property (RESILIENT cluster, low vacancy)

### Negotiated Move-Out (Cash for Keys)

Often the most cost-effective resolution for deep delinquency:
- Offer tenant $500-$1,500 to vacate voluntarily within 7-14 days
- Tenant signs a release and move-out agreement
- Avoids 2-4 months of legal process and legal fees
- Works best for COH tenants and short-tenure TOH tenants
- **Total cost:** $500-$1,500 offer + $0 legal = far less than eviction
- **Caution:** Document everything. Have the tenant sign a written agreement that they are leaving voluntarily.

---

## Seasonal Adjustment Methodology

### Monthly Risk Adjustments

Based on CFPB research, industry data, and Sunrise portfolio historical patterns:

| Month | Adjustment | Driver | Evidence |
|-------|------------|--------|----------|
| **January** | +5 | Holiday spending hangover, heating costs peak | CFPB: late fees spike in Q1 |
| **February** | +3 | Continued winter stress, tax refunds pending | Income tax refund anticipation |
| **March** | -3 | Tax refunds arrive (EITC, CTC) | CFPB: late fees drop in spring |
| **April** | -2 | Tax refund spending continues | Seasonal improvement |
| **May** | 0 | Baseline | Stable period |
| **June** | 0 | Baseline | Stable period |
| **July** | +3 | Back-to-school expenses begin | Family stress, childcare costs |
| **August** | +3 | Back-to-school peak, summer utility bills | Seasonal stress |
| **September** | 0 | Baseline | Stabilization |
| **October** | 0 | Baseline | Pre-holiday calm |
| **November** | +3 | Holiday spending begins, heating costs rise | Discretionary spending competes |
| **December** | +5 | Holiday spending peak, year-end stress | Highest delinquency risk |

**IMPORTANT: Post-COVID shift.** CFPB data (2025) noted that the traditional spring tax-refund dip in late payments is weakening. "Late payments did not ease in spring tax-refund season as expected, suggesting deeper cash flow misalignment." Monitor whether March/April adjustments should be reduced.

### Implementation Note

The v2.0 code does NOT currently implement monthly seasonal adjustments directly. The seasonal effect is partially captured by the economic state modifier (SLOWDOWN/RECESSION periods often coincide with seasonal stress). A future v3.0 enhancement should add explicit monthly adjustments.

**To add seasonal adjustments to the model:**
```python
SEASONAL_ADJUSTMENTS = {
    1: 5, 2: 3, 3: -3, 4: -2, 5: 0, 6: 0,
    7: 3, 8: 3, 9: 0, 10: 0, 11: 3, 12: 5
}
# Apply in calculate_all_risk_scores():
seasonal_adj = SEASONAL_ADJUSTMENTS.get(datetime.now().month, 0)
total_score = base_score + state_modifier + seasonal_adj
```

---

## autoMHatic Referral Integration

### Overview

autoMHatic Financial is Sunrise's lending partner for resident home improvement loans. The delinquency risk model and the loan candidate screening model (`loan_candidate_screening.py`) are INVERSE systems:
- **Delinquency model:** Identifies tenants likely to MISS payments
- **Loan screening:** Identifies tenants with STRONG payment history for loan referrals

### Eligibility Criteria for autoMHatic Referral

From `loan_candidate_screening.py`:

| Criterion | Requirement | Rationale |
|-----------|------------|-----------|
| Consecutive on-time payments | 9+ months | Demonstrates payment reliability |
| Current balance | $0 or fully current | No outstanding delinquency |
| Tenure | 9+ months | Established residency |
| Account status | Active/Current | Not in eviction or move-out process |
| Home ownership | TOH preferred, COH eligible | Loan secured against home improvements |

### Integration with Delinquency Model

**Cross-referencing the two models:**
- A tenant in LOW risk tier (0-30) with 9+ months of on-time payments is a strong autoMHatic candidate
- A tenant who was previously HIGH but has completed a payment plan and maintained 9+ months of on-time payments SINCE resolution is a rehabilitation success story — consider for referral with a note about their history
- Never refer a tenant in MEDIUM, HIGH, or CRITICAL tier
- If a previously referred tenant moves to ELEVATED/HIGH, notify autoMHatic immediately

### Conversion Tracking

Track these metrics quarterly:
- Number of tenants meeting eligibility criteria (potential referral pool)
- Number actually referred to autoMHatic
- Number who completed a loan application
- Number approved and funded
- Default rate on autoMHatic loans (should be < 3% given screening criteria)
- Impact on lot rent retention (do loan recipients stay longer?)

### Quarterly Report Output

The loan candidate screening script outputs to Google Sheets with:
- Candidate list (name, property, tenure, payment streak, home ownership type)
- Summary statistics by property
- Trend analysis (is the eligible pool growing or shrinking?)

---

## Bad Debt Reserve Calculation

### Methodology: Aging-Based Reserve

Sunrise should use the **accounts receivable aging method** for bad debt reserves, aligned with GAAP and standard property management accounting.

| Aging Bucket | Reserve % | Rationale |
|-------------|-----------|-----------|
| Current (0-30 days) | 2% | Minor risk, most will pay |
| 31-60 days | 10% | Moderate risk, some will cure |
| 61-90 days | 25% | Significant risk, escalating |
| 91-120 days | 50% | High risk, many will not pay |
| 120+ days | 75-90% | Near-certain loss |
| Post-move-out balances | 95% | Collection rate is extremely low |

### Calculation Process (Monthly)

```
1. Pull DelinquencyAging report from RentManager (by tenant)
2. Classify each balance into aging buckets
3. Apply reserve percentages
4. Sum = Required Bad Debt Reserve

Example:
  Current (0-30):    $50,000 × 2%  = $1,000
  31-60 days:        $15,000 × 10% = $1,500
  61-90 days:         $8,000 × 25% = $2,000
  91-120 days:        $5,000 × 50% = $2,500
  120+ days:          $3,000 × 80% = $2,400
  Post-move-out:      $4,000 × 95% = $3,800
  ─────────────────────────────────────────
  Total AR:          $85,000
  Required Reserve:  $13,200 (15.5% of total AR)
```

### Reserve Adequacy Testing (Quarterly)

Compare actual write-offs to reserved amounts:
- If actual write-offs consistently EXCEED reserves: increase reserve percentages
- If actual write-offs are consistently BELOW reserves: consider decreasing (but err conservative)
- Track the **coverage ratio:** Reserve ÷ Actual Write-offs. Target: 1.0x-1.2x

### Portfolio-Level Bad Debt Target

- **Target bad debt expense:** < 1.5% of gross rental revenue
- **Industry benchmark (MHPs):** 1-2% bad debt is typical for well-managed communities
- **MHP sector delinquency rate (CMBS):** 1.39% as of mid-2025, far below multifamily (6.19%) and office (10.99%)
- **Sunrise target:** < 3% delinquency rate, < 1.5% bad debt write-off

---

## Model Performance Monitoring

### Accuracy Metrics

**Monthly validation:**
1. **Prediction accuracy:** What percentage of CRITICAL-tier tenants actually defaulted (missed 2+ payments or were evicted) within 90 days?
2. **False positive rate:** What percentage of HIGH/CRITICAL tenants resolved without intervention (i.e., would have paid anyway)?
3. **False negative rate:** What percentage of tenants who defaulted were scored LOW or MEDIUM in the prior month's run?
4. **Tier stability:** What percentage of tenants stayed in the same tier month-over-month? (High churn = noisy model)

**Target metrics:**
| Metric | Target | Concern Threshold |
|--------|--------|-------------------|
| CRITICAL prediction accuracy (90-day) | > 60% | < 40% |
| False positive rate (HIGH+CRITICAL) | < 30% | > 50% |
| False negative rate | < 10% | > 20% |
| Tier stability (LOW tier) | > 90% | < 80% |

### Recalibration Triggers

Recalibrate the model (adjust weights, thresholds, or scoring logic) when:
- False negative rate exceeds 15% for 2 consecutive months
- Tier distribution deviates >10 percentage points from target (e.g., CRITICAL at 13% instead of 3%)
- Economic regime change (EXPANSION → RECESSION) — verify that the state modifier is adjusting distribution appropriately
- Portfolio composition changes significantly (new property acquisition, disposition)
- Post-mortem on a large unexpected write-off reveals a scoring gap

### Backtesting Protocol

Quarterly:
1. Take the risk scores from 3 months ago
2. Compare to actual outcomes (who paid, who defaulted, who left)
3. Calculate confusion matrix (true positive, false positive, true negative, false negative)
4. Compute precision, recall, and F1 score for each tier
5. Document findings in `/Operations/delinquency-risk-model/backtest/` directory

### Future Enhancements (ML Upgrade Path)

The current model is a **rules-based scoring system** — interpretable, auditable, and aligned with Fair Housing requirements. If the portfolio grows significantly, consider:

1. **Logistic regression:** Add interaction terms (tenure × payment history, cluster × economic state). Still interpretable and auditable.
2. **Decision trees / random forests:** Better at capturing non-linear patterns. Require SHAP or LIME for explainability (Fair Housing).
3. **XGBoost / gradient boosting:** Highest predictive accuracy, but "black box" risk. Would need robust disparate impact testing before deployment.
4. **Feature expansion:** Add utility payment data (if community-billed), maintenance request frequency, communication responsiveness score, local crime/economic data.

**Constraint:** Any ML model used for tenant-facing decisions MUST be explainable. HUD guidance and the Fair Housing Act's disparate impact standard require the ability to explain WHY a tenant received a particular score. Stick with rules-based or logistic regression unless legal counsel approves a more complex model with adequate explainability tooling.

---

## Ethical AI & Fair Housing Guardrails

### Disparate Impact Testing

*Last verified: March 2026. Informed by HUD guidance on algorithms and AI in tenant screening (May 2024) and Texas Dept. of Housing v. Inclusive Communities Project (2015).*

**The legal standard:** Even a facially neutral algorithm can violate the Fair Housing Act if it has a disproportionate adverse effect on a protected class. The plaintiff need not prove discriminatory intent — only statistical impact and a causal connection to the model's criteria.

**Required testing (quarterly):**

1. **Demographic parity test:** Do risk tier distributions differ significantly across racial/ethnic groups at each property?
   - **Challenge:** Sunrise does not (and should not) collect race/ethnicity data on tenants. Use census tract demographics as a proxy.
   - Compare risk tier distributions for properties in majority-minority census tracts vs. majority-white tracts.
   - If HIGH/CRITICAL rates are >1.5x higher in majority-minority tracts after controlling for economic factors, investigate.

2. **Feature audit:** Review each scoring component for proxy discrimination potential.
   - **Payment history:** Generally defensible — directly measures financial behavior.
   - **Tenure:** Low risk for bias — newer tenants are higher risk regardless of demographics.
   - **Economic factors (unemployment):** CAUTION — unemployment rates correlate with race at the geographic level. The model uses STATE-level rates (coarser grain = less proxy risk). Do NOT use zip-code-level unemployment without disparate impact testing.
   - **Property cluster:** CAUTION — if SENSITIVE properties correlate with majority-minority communities, this feature could create disparate impact. Verify.
   - **Balance aging:** Generally defensible — measures actual financial position.

3. **Outcome audit:** Do eviction rates differ across protected classes? Track and report annually.

### Bias Mitigation Strategies

1. **Use the model for INTERVENTION, not EXCLUSION.** The delinquency risk model determines which tenants get MORE help (payment plans, outreach), not which tenants are denied housing. This framing is legally and ethically stronger.
2. **Human-in-the-loop for all adverse actions.** No eviction should proceed based solely on a risk score. A property manager must review the tenant's specific circumstances.
3. **Document all exceptions.** If a manager overrides the model's recommendation (e.g., offers a payment plan to a CRITICAL tenant, or does NOT pursue a HIGH-balance LOW-tier tenant), document the business justification.
4. **Annual disparate impact report.** Even if no issues are found, the fact of testing demonstrates good faith.
5. **Model transparency.** The scoring methodology should be available to tenants upon request. A tenant should be able to understand why they received their score.

### What the Model Should NEVER Be Used For

- Screening prospective tenants for admission (use standard screening criteria: credit, background, income)
- Determining rent amounts or rent increases
- Selecting tenants for lease non-renewal (without independent business justification)
- Any purpose where the tenant has no opportunity to cure or respond
- Sharing risk scores with third parties (other landlords, credit bureaus) without proper authorization

### Documentation Requirements

For every adverse action (demand letter, eviction filing, reporting to collections):
1. Risk score at time of action
2. Specific delinquent balance and aging
3. Collection attempts made (dates, methods, outcomes)
4. Payment plan offered (yes/no, terms, response)
5. Business justification for the specific action taken
6. Manager signature and date

---

## Decision Trees

### Decision Tree 1: Intervention Type Selection

```
Tenant flagged HIGH or CRITICAL
│
├─ Is tenant communicative? (responded to 1+ contact attempt)
│   ├─ YES → Has tenant experienced documented hardship?
│   │   ├─ YES → Offer Scenario A payment plan (hardship)
│   │   │   └─ Refer to tenant assistance programs (see below)
│   │   └─ NO → Is this a chronic lateness pattern?
│   │       ├─ YES → Offer Scenario B (auto-pay transition)
│   │       └─ NO → Offer Scenario C (structured paydown)
│   │
│   └─ NO → Has there been 3+ contact attempts across 2+ channels?
│       ├─ NO → Continue contact attempts (phone, text, door visit, mail)
│       └─ YES → Is the tenant still residing at the property?
│           ├─ YES → Formal demand letter (certified mail)
│           │   └─ 7 days no response → Legal consultation
│           └─ NO (skip) → Skip trace, report to collections agency
```

### Decision Tree 2: Eviction vs. Payment Plan

```
CRITICAL tenant, 60+ days past due
│
├─ Is balance < estimated eviction cost?
│   ├─ YES → Payment plan is financially superior
│   │   └─ Offer payment plan → If accepted → Monitor
│   │                        → If refused → Cash-for-keys offer
│   │                        → If no response → Proceed to eviction
│   └─ NO → Does the tenant have prior payment plan default?
│       ├─ YES → Proceed to eviction (pattern of non-compliance)
│       └─ NO → One final payment plan offer (short timeline, 50% down)
│           └─ Accepted → Monitor strictly (one miss = eviction)
│           └─ Refused → Proceed to eviction
│
├─ Special circumstances check:
│   ├─ Bankruptcy filed? → STOP all collection, consult counsel
│   ├─ SCRA active duty? → STOP, court order required
│   ├─ Disability accommodation request? → Engage in interactive process
│   └─ Domestic violence situation? → Follow state protections (many states
│       prohibit eviction of DV victims for lease violations caused by abuser)
```

### Decision Tree 3: Write-Off Timing

```
Tenant balance 120+ days past due
│
├─ Is tenant still in the unit?
│   ├─ YES → Continue collection efforts (eviction if warranted)
│   │   └─ Do NOT write off while tenant is in the unit
│   └─ NO → Has the tenant moved out?
│       ├─ YES → How long ago?
│       │   ├─ < 6 months → Send to collections agency, maintain on books
│       │   ├─ 6-12 months → Write off 75%, send to collections
│       │   └─ 12+ months → Write off 95%, minimal collection effort
│       └─ UNKNOWN (skip) → Skip trace, then treat as moved out
│
├─ Write-off approval:
│   ├─ < $500 → Property manager can approve
│   ├─ $500-$2,000 → Regional manager approval
│   └─ $2,000+ → VP/Controller approval
```

### Tenant Assistance Program Referrals

When a tenant's delinquency stems from a documented hardship, refer to:

| Resource | Type | Coverage |
|----------|------|----------|
| **211 Hotline** | National | Connects to local assistance programs |
| **Salvation Army** | Nonprofit | Emergency rent/utility assistance |
| **Catholic Charities** | Nonprofit | Emergency assistance (not faith-restricted) |
| **United Way** | Nonprofit | Community-specific emergency funds |
| **Local Community Action Agencies** | Government | CSBG, ESG, TBRA programs |
| **State ERA successors** | Government | Post-COVID state/county rental assistance funds |
| **VA (if veteran)** | Government | SSVF (Supportive Services for Veteran Families) |
| **HUD.gov/FindShelter** | Government | National resource finder |

**Note:** Federal ERA (Emergency Rental Assistance) programs ended Sept 2025, but many states repurposed remaining funds into successor programs. The landscape is fragmented — 211 is the best single entry point.

---

## Common Gotchas

### 1. Fair Housing in Automated Scoring
**Gotcha:** Using zip-code-level economic data (unemployment, crime, income) as model features can create proxy discrimination. Even without intent, if SENSITIVE properties correlate with majority-minority communities, the cluster feature amplifies racial disparities.
**Fix:** Use state-level economic data (coarser grain) and test for disparate impact quarterly. The current model uses state-level BLS data, which is safer.

### 2. Model Drift
**Gotcha:** Scoring thresholds calibrated during EXPANSION may produce wildly different tier distributions during RECESSION. If 25% of tenants suddenly score CRITICAL, the action protocols become unworkable (site managers cannot call 25% of tenants within 24 hours).
**Fix:** The economic state modifier is designed to handle this gradually. But monitor tier distribution monthly. If CRITICAL exceeds 8%, consider temporarily raising the CRITICAL threshold or prioritizing by balance amount.

### 3. Seasonal False Positives
**Gotcha:** January and December seasonal adjustments (+5) can push borderline MEDIUM tenants into HIGH, triggering unnecessary intervention calls. These tenants often pay within 10-15 days.
**Fix:** For tenants who score HIGH only because of seasonal adjustment (base score is MEDIUM), delay intervention by 5 days. If they pay within that window, suppress the alert.

### 4. New Tenant Penalty
**Gotcha:** Tenants with <6 months tenure get 10 tenure points AND 20 payment history points (no history = medium risk). Combined, that is 30 points — already at the MEDIUM/HIGH boundary before any actual delinquency.
**Fix:** For tenants with <3 months of data, suppress risk tier to LOW unless they have actual delinquent balances. The model should not flag brand-new tenants who have not yet had the opportunity to build a payment history.

### 5. TOH Abandoned Homes
**Gotcha:** When a TOH resident abandons their home (stops paying, leaves without notice), the balance grows but there is no one to collect from. The risk model scores them CRITICAL indefinitely.
**Fix:** Reclassify abandoned TOH lots to unit type "ABN" (81) or "TOH-ABAN" (98). The model already excludes "OTHER" unit types from scoring. Process the abandonment through legal channels (lien, title transfer, demo).

### 6. Deceased Residents
**Gotcha:** When a resident dies, their account may go delinquent. The risk model treats this as a normal delinquency.
**Fix:** Tag deceased residents immediately in RentManager (status change). The model filters on `Status == "Current"` — changing status removes them from scoring. Work with the estate or next of kin separately from normal collections.

### 7. Seasonal + Economic Double-Counting
**Gotcha:** During a RECESSION in January, a tenant gets +10 (state modifier) + +5 (seasonal) = +15 points from non-behavioral factors alone. This can push stable tenants into HIGH tier.
**Fix:** Cap the combined seasonal + state modifier at +12 points. Or weight seasonal adjustments by economic state (reduce seasonal impact during RECESSION since the state modifier already captures the macro stress).

### 8. Data Freshness
**Gotcha:** The model runs on CSV exports from RentManager. If exports are stale (not pulled this month), scores reflect outdated data.
**Fix:** The model auto-detects the most recent YYYY-MM folder in the reports directory. Always run the RentManager exports BEFORE running the risk model. Add a data freshness check: if the most recent export is >45 days old, print a warning and refuse to run.

### 9. Action List Balance Threshold
**Gotcha:** The Action List tab filters for HIGH/CRITICAL tenants with ≥$200 delinquent balance. Tenants scoring HIGH on behavioral factors but with $0 balance (they paid late but eventually paid) are excluded from the action list but still scored HIGH.
**Fix:** This is intentional — $0 balance tenants do not need immediate collection action. But their HIGH score signals they are AT RISK for future delinquency. Property managers should still be aware of these tenants through the full Risk_Scores tab.

### 10. Multiple Leases per Tenant
**Gotcha:** Some tenants have multiple lease records in RentManager (renewals, transfers). The model uses `sort_values("MoveInDate", ascending=False)` and takes the first lease per tenant. If the most recent lease has incorrect data, it propagates.
**Fix:** When investigating an individual tenant's score, always verify the MoveInDate and UnitID against the actual lease in RentManager.

---

## Output Formats

### Individual Tenant Risk Report
```markdown
## Tenant Risk Assessment: [Name]

**Property:** [Community ShortName]
**Unit/Lot:** [Unit Name]
**Move-In Date:** [YYYY-MM-DD]
**Home Ownership:** [TOH/COH]
**Current Balance:** $[TotalBalance]
**Delinquent Balance:** $[DelinquentBalance]

### Risk Score: [XX]/100 — [TIER]

| Component | Score | Max | Notes |
|-----------|-------|-----|-------|
| Payment History | X | 35 | [Late count, trend] |
| Balance Aging | X | 25 | [Aging bucket] |
| Economic | X | 15 | [State, unemployment rate] |
| Property Cluster | X | 10 | [Cluster type] |
| Tenure | X | 10 | [Years] |
| Lease Expiration | X | 5 | [Days to renewal] |
| State Modifier | X | 10 | [Economic state] |

### Key Risk Factors
- [Factor 1: e.g., "6 late payments in past 12 months"]
- [Factor 2: e.g., "Balance 61-90 days past due"]
- [Factor 3: e.g., "Sensitive property during SLOWDOWN"]

### Recommended Action
[Specific intervention from tier protocol]

### Payment History (Last 12 Months)
| Month | Due Date | Paid Date | Days Late | Amount |
|-------|----------|-----------|-----------|--------|
| | | | | |

### Collection Activity Log
| Date | Action | Outcome | Next Step |
|------|--------|---------|-----------|
| | | | |
```

### Portfolio Risk Summary
```markdown
## Delinquency Risk Summary: [YYYY-MM-DD]

### Risk Distribution
| Tier | Count | % of Portfolio | Total Delinquent Balance | Avg Score |
|------|-------|----------------|--------------------------|-----------|
| CRITICAL | X | Y% | $Z | XX |
| HIGH | X | Y% | $Z | XX |
| MEDIUM | X | Y% | $Z | XX |
| LOW | X | Y% | $Z | XX |
| **TOTAL** | **X** | **100%** | **$Z** | **XX** |

### vs. Targets
| Metric | Actual | Target | Status |
|--------|--------|--------|--------|
| Portfolio delinquency rate | X% | < 3% | [ON TRACK / AT RISK / OFF TRACK] |
| Collection rate | X% | > 98% | [ON TRACK / AT RISK / OFF TRACK] |
| Bad debt % of revenue | X% | < 1.5% | [ON TRACK / AT RISK / OFF TRACK] |

### Month-over-Month Trend
- New CRITICAL this month: X
- Moved to CRITICAL from HIGH: X
- Resolved (CRITICAL/HIGH → MEDIUM/LOW): X
- Net change in CRITICAL+HIGH: +/- X

### Property Concentration
| Property | Cluster | CRITICAL | HIGH | % HIGH+ | Total Delinquent |
|----------|---------|----------|------|---------|-----------------|
| [Sorted by % HIGH+ descending] | | | | | |

### Top 10 Action Items
| # | Tenant | Property | Score | Tier | Balance | Recommended Action |
|---|--------|----------|-------|------|---------|-------------------|
| 1 | | | | | | |
| ... | | | | | | |

### Economic Context
- Economic State: [EXPANSION/SLOWDOWN/etc.]
- State Modifier: +[X] points
- Unemployment rates: [State-by-state summary]
```

### Collection Activity Report (Weekly)
```markdown
## Collection Activity Report: Week of [Date]

### Summary
| Metric | This Week | MTD | Target |
|--------|-----------|-----|--------|
| Contact attempts made | X | X | |
| Contacts reached | X | X | |
| Payment plans established | X | X | |
| Payment plans in compliance | X | X | |
| Demand letters sent | X | X | |
| Eviction filings | X | X | |
| Balances resolved | $X | $X | |

### Active Payment Plans
| Tenant | Property | Original Balance | Remaining | Next Installment | Status |
|--------|----------|-----------------|-----------|-----------------|--------|
| | | | | | [On Track / Missed / Complete] |

### Escalations This Week
| Tenant | Property | From Tier | To Tier | Balance | Action Taken |
|--------|----------|-----------|---------|---------|-------------|
| | | | | | |
```

---

## References

### Internal Resources
| Resource | Location |
|----------|----------|
| Delinquency Risk Model (code) | `/Operations/delinquency-risk-model/delinquency_risk_model.py` |
| Loan Candidate Screening | `/Operations/loan-candidate-screening/loan_candidate_screening.py` |
| Probability Engine (clusters) | `/Asset Management/probability-engine/` |
| Data Dictionary | `/Operations/data-pipeline-consolidation/DATA_DICTIONARY.md` |
| Risk Dashboard (Google Sheets) | Sheet ID: `1kFhhIqICSg-fpUYMdYjgFLm9p7PMa97SbCIy8WHDxUM` |
| RentManager MCP | `/rentmanager-mcp/` |
| Monthly KPI Export | `/rentmanager-mcp/export_monthly_kpi_dashboard.py` |

### External Sources & Research
| Source | Topic | URL |
|--------|-------|-----|
| CFPB Rental Payment Data | Seasonal delinquency, late fee trends | consumerfinance.gov/data-research/research-reports/ |
| BLS LAUS | State unemployment rates | api.bls.gov/publicAPI/v2/timeseries/data/ |
| Nolo Legal Encyclopedia | State eviction laws by state | nolo.com/legal-encyclopedia/ |
| iPropertyManagement | Eviction process timelines | ipropertymanagement.com/laws/eviction-process |
| Snappt / Innago | Eviction cost analysis | snappt.com/blog/eviction-cost/ |
| HUD | Fair Housing Act guidance on algorithms and AI | consumerfinancialserviceslawmonitor.com (May 2024 guidance) |
| Georgetown Poverty Journal | Discriminatory impacts of AI tenant screening | law.georgetown.edu/poverty-journal/ |
| TX Dept of Housing v. Inclusive Communities (2015) | Disparate impact standard | Supreme Court |
| MBA Delinquency Reports | CMBS delinquency benchmarks by asset class | mba.org |
| NMHC Rent Payment Tracker | National rent payment timing data | nmhc.org |
| Treasury ERA Program | Emergency Rental Assistance status | home.treasury.gov |
| DOD SCRA Verification | Servicemember status check | scra.dmdc.osd.mil |

### Industry Benchmarks
| Metric | MHP Benchmark | Source |
|--------|--------------|-------|
| CMBS delinquency rate (MH) | 1.39% (mid-2025) | MBA/Trepp |
| CMBS delinquency rate (multifamily) | 6.19% (mid-2025) | MBA |
| CMBS delinquency rate (office) | 10.99% (mid-2025) | MBA |
| Renter late fee incidence | 14% (Nov 2024) | CFPB |
| Average eviction cost (total) | $3,500-$10,000 | Snappt/Innago |
| Average MHP resident tenure | 10-14 years | Industry surveys |
| Average apartment resident tenure | 12-16 months | NAA |

### Legal References (by State)
| State | Statute | Key Provision |
|-------|---------|---------------|
| OH | ORC §1923 | 3-day notice, forcible entry and detainer |
| IN | IC §32-31-3 | 10-day notice with cure right |
| MD | Real Property §8-402 | 10-day notice; rent escrow available |
| MI | MCL §600.5714 | 7-day demand for possession |
| FL | F.S. §83.56 | 3-day notice; 90-day rent increase notice for MH |
| WV | WV Code §37-6-5 | 5-day notice with cure right |
| PA | 68 P.S. §250.501 | 10-day notice, no automatic cure |
| AL | Code §35-9A-421 | 7 business days with cure right |
| GA | O.C.G.A. §44-7-50 | No notice required for nonpayment |
| AR | A.C.A. §18-60-304 | 3-day notice (if 5+ days past due) |
| MN | M.S. §504B.135 | 14-day notice (at-will); 90-day rent increase for MH |
| ND | NDCC §33-06-01 | 3-day notice |
| WI | Wis. Stat. §704.17 | 5-day (short lease) / 30-day (long lease) |
| MO | RSMo §441.060 | 5-day notice, no cure right |
| IL | 735 ILCS 5/9-209 | 5-day notice with cure right |

*All legal references last verified: March 2026. These are general summaries — manufactured housing may have additional state-specific protections. Always consult legal counsel before initiating eviction proceedings.*