# Placeholder Variables & Future Implementation **Date**: 2025-10-27 **Status**: Pipeline working with placeholders for single-snapshot Company data --- ## ⚠️ Variables Using Placeholders or Proxies ### 🔴 HIGH PRIORITY #### **`survival` (Currently: Proxy)** - **Current Implementation**: `survival = later_success` (from LastFinancingDealType) - **Proper Implementation**: ```python survival = 1 if (company in 20230501 snapshot) AND (LastFinancingDate >= 2021-11-01) ``` - **Data Needed**: Deal20211201.dat, Deal20220101.dat, Deal20230501.dat - **Impact**: HIGH - This is the main DV for H2 Main model - **Current Limitation**: Proxy may misclassify survival (e.g., got Series B in 2019 but failed in 2022) --- ### 🟡 MEDIUM PRIORITY #### **`series_a_funding` (Currently: Proxy)** - **Current**: `series_a_funding = early_funding_musd` (FirstFinancingSize) - **Proper**: Extract from Deal data where `VCRound == 'Series A'` - **Data Needed**: Deal data with VCRound column - **Impact**: MEDIUM - Needed for H2 Robustness control variable #### **`series_b_funding` (Currently: Missing)** - **Current**: `series_b_funding = NaN` (all missing) - **Proper**: Extract from Deal data where `VCRound == 'Series B'` - **Data Needed**: Deal data with VCRound column - **Impact**: MEDIUM - This is the DV for H2 Robustness model - **Current Limitation**: H2 Robustness test cannot run --- ### 🟢 LOW PRIORITY #### **`founder_credibility` (Currently: Placeholder)** - **Current**: `founder_credibility = 0` (all companies) - **Future Implementation**: ```python founder_credibility = ( prior_exits * 0.3 + log(patents + 1) * 0.2 + advanced_degree * 0.2 + prior_companies * 0.2 + awards * 0.1 ) ``` - **Data Needed**: Founder profile data (not in Company.dat) - **Impact**: LOW - Results may be significant without it (other controls compensate) #### **`is_down_round` (Currently: Placeholder)** - **Current**: `is_down_round = 0` (no down rounds detected) - **Proper**: Compare PostValuation across rounds per company - **Data Needed**: Deal data with PostValuation and DealDate - **Impact**: LOW - Optional control for H2 Robustness only #### **`sector_fe` (Currently: Working with limitations)** - **Current**: 8 categories via keyword matching - **Improvement**: Manual validation, better industry taxonomy, or PitchBook's PrimaryIndustry - **Impact**: LOW - Current categorization sufficient, can be refined #### **`vagueness` (Currently: Working)** - **Current**: Keyword counting (vague vs precise words) - **Enhancement**: LLM-based scoring, context-aware analysis - **Impact**: LOW - Current method established in literature --- ## 📋 Implementation Priority ### **Phase 1: Get True Survival (After Presentation)** 🔴 1. Obtain Deal data files (3 snapshots) 2. Implement `create_survival_from_snapshots()` with Deal data 3. Modify `run_pipeline.py` to load Company + Deal data together 4. Re-run analysis with true 18-month survival window 5. Compare results to proxy-based results ### **Phase 2: Enable H2 Robustness Test** 🟡 1. Extract Series A and Series B funding from Deal data 2. Implement down round detection 3. Run H2 Robustness: `series_b_funding ~ vagueness * integration_cost + series_a_funding + controls` 4. Compare to H2 Main results ### **Phase 3: Refine Variables** 🟢 1. Implement founder_credibility if data available (optional) 2. Validate sector categorization accuracy 3. Consider LLM-based vagueness scoring (optional) --- ## 🎯 For Today's Presentation ### **What Works** ✅ - ✅ Vagueness calculation (keyword-based, 0-100 scale) - ✅ H1 test: Early Funding ~ Vagueness + Controls - ✅ H2 Main test: Survival(proxy) ~ Vagueness × Integration Cost - ✅ Integration cost classification (hardware vs software) - ✅ Sector fixed effects (8 categories) - ✅ All visualizations (scatter, interaction, ROC, coefficient comparison) ### **What's Proxied** ⚠️ - ⚠️ Survival uses later_success (need Deal data for true 18-month window) - ⚠️ Series A funding uses FirstFinancingSize (need Deal data for precise round) - ⚠️ Founder credibility = 0 (low impact on results) ### **What's Missing** ❌ - ❌ H2 Robustness test (needs Series B funding from Deal data) - ❌ Down round detection (needs Deal data) ### **Transparency Statement for Professors** > "Our analysis uses the May 2023 Company snapshot (504K companies). The **survival variable** currently uses 'later success' (reached Series B+) as a proxy. The proper 18-month survival window requires Deal data with financing dates, which we'll implement in the next iteration. Results should be directionally correct, though the proxy may introduce some misclassification. **Founder credibility** is a placeholder (=0) as we don't have founder profile data, but other controls (firm size, age, sector) should compensate." --- ## 🔬 Technical Details ### **Current Data Flow** ``` Company20230501.dat (631 MB, 504K rows) ↓ Feature Engineering ├─ vagueness (keyword count) ✅ ├─ integration_cost (keyword classification) ✅ ├─ early_funding (FirstFinancingSize) ✅ ├─ sector_fe (8 categories from Keywords) ✅ ├─ founder_credibility = 0 ⚠️ └─ survival = later_success ⚠️ ↓ Hypothesis Tests ├─ H1: Early Funding ~ Vagueness ✅ ├─ H2 Main: Survival(proxy) ~ Vagueness × Integration ✅ └─ H2 Robustness: SKIPPED ❌ (needs Deal data) ↓ Visualizations ✅ ``` ### **Future Data Flow (with Deal Data)** ``` Company20230501.dat + Deal{20211201,20220101,20230501}.dat ↓ Feature Engineering ├─ TRUE survival (3-snapshot + LastFinancingDate logic) ✅ ├─ series_a_funding (from VCRound='Series A') ✅ ├─ series_b_funding (from VCRound='Series B') ✅ └─ is_down_round (PostValuation comparison) ✅ ↓ Hypothesis Tests ├─ H1: Early Funding ~ Vagueness ✅ ├─ H2 Main: TRUE Survival ~ Vagueness × Integration ✅ └─ H2 Robustness: Series B ~ Vagueness × Integration + Series A ✅ ``` --- ## 📞 Questions? Review: 1. `IMPLEMENTATION_COMPLETE.md` - Full implementation details 2. `PHASE1_MAPPING_AND_GAPS.md` - Design decisions 3. This file - Placeholder status and future work **Ready for next iteration?** Get Deal data files and implement Phase 1 (True Survival).