Forecasting India→US Export Loss & Recovery Under U.S. Tariffs

What this project delivers

• A baseline forecast of FY2025 exports (no tariffs) by category using XGBoost (XGBRegressor)
• A tariff shock applied via price elasticity of demand
• A lever stack (rebates, cost cuts, demand uplift) that absorbs part of the shock and shows how much loss is avoided
• A worked example for Textiles with explicit $ and % impacts

---

1) Data acquisition & preparation

Data source (9 years):

• India’s Ministry of Commerce / DGCIS (annual exports by broad category)

Schema used

Field	Example	Notes
Year	2016…2024	Fiscal years
Export Category	Textile Products, Pharma, Engineering, etc.	9 aggregate buckets
Export Value (USD Million)	10,959	Annual value

Pre‑processing

• Sort by Export Category, Year
• Create year_index = Year - Year.min()
• One‑hot encode Export Category for modeling
• (Optional, if monthly data) create lags and rolling means per category

---

2) Model selection & why XGBoost

Why XGBoost / XGBRegressor

• Handles tabular data with non‑linearities
• Works well with sparse one‑hot category features
• Fast to train; robust to moderate feature scaling issues
• Allows explainability (feature importance, SHAP)

Model object

from xgboost import XGBRegressor
model = XGBRegressor(
    n_estimators=300,
    max_depth=4,
    learning_rate=0.05,
    subsample=0.9,
    colsample_bytree=0.9,
    random_state=42
)

Training setup

• Features: year_index + one‑hot(Export Category)
• Target: Export Value (USD Million)
• Train window: 2016–2024
• Predict: 2025 baseline (no tariff)

---

3) The dynamics of XGBoost & XGBRegressor (short)

• XGBoost builds additive trees optimizing a regularized objective (squared error for regression)
• learning_rate shrinks each tree’s contribution; max_depth controls interaction order; subsample and colsample_bytree add randomness to reduce overfit
• Feature importance: gain/weight/cover or SHAP values to see which features drive forecast (typically year_index + category dummies)

---

4) Elasticity concept (what, why, how)

Definition

$E = \frac{\%\Delta \text{Quantity}}{\%\Delta \text{Price}}$

A price increase from tariffs reduces quantity (E is typically negative).

How assigned in this project (assumptions guided by literature & market structure)

Export Category	Elasticity (E)	Rationale (short)
Textile Products	−1.1	Highly substitutable, price‑competitive
Pharma Products	−0.3	Essential demand, regulated
Gems & Jewellery	−1.0	Discretionary, deferrable
Engineering Products	−1.2	Cost‑driven B2B procurement
Chemicals & Allied	−0.9	Input market with alternatives
Electronics / E&SW	−0.8	Competitive, partial brand pull
Marine	−0.7	Moderate substitutability
Leather	−1.0	Fashion/discretionary
Agricultural	−0.4	Semi‑essential, partial substitutes

How elasticity is measured in the real world (if you extend the study)

• Panel regression on (log) quantity vs (log) price with fixed effects for product & market, controlling for income, FX, and policy dummies
• Event studies around policy/tariff changes in comparable markets
• Meta‑analyses from trade orgs (WITS/IMF/UNCTAD) to benchmark category ranges

---

5) From forecast to tariff shock

Step 1: Baseline (no tariff): Use XGBoost to predict 2025 without tariff.

Step 2: Tariff‑only scenario: Tariff rate T (e.g., 50%) → if fully passed to price, price change = +T. Quantity effect:

$$ %\Delta Q = E \times %\Delta P $$

Value proxy ≈ Baseline × (1 + E × T) (for small changes) or combine with ASP assumptions if available.

---

6) Absorption framework & levers

Two different knobs

• absorb: share of tariff not passed to buyer (margin squeeze). net_tariff_to_price = T × (1 − absorb)
• pass_to_price: share of internal savings passed down to the buyer (price relief). price_offset_from_savings = pass_to_price × cost_savings_total

Effective price change (after levers)

$\Delta P_{\text{eff}} = \underbrace{T(1- \text{absorb})}_{\text{net tariff}} \;-\; \underbrace{\text{rebate\_rate}}_{\text{RoSCTL/RoDTEP}} \;-\; \underbrace{\text{pass\_to\_price}\times \text{cost\_savings\_total}}_{\text{shared savings}}$

Quantity change

$\Delta Q\% = E \times \Delta P_{\text{eff}}$

Demand‑side uplift (loyalty, bundling, marketing, customization, service) applied multiplicatively to volume: (1 + demand_uplift).

Quantified levers & benchmarks used

• Rebates:

  • RoSCTL (textiles): 6.05%–8.2% (used ~7% in base)
  • RoDTEP (many lines in chemicals/pharma/iron & steel): ~0.5%–2% (commonly 0.8%)

• Cost savings (COGS/logistics/procurement):

  • Vendor rationalization: 3–5%
  • Logistics optimization / FTA routing: 8–15%
  • Better commodity forecasting: 5–10%
  • In the model we sum these (e.g., ~21%) and pass 40–60% to price (e.g., 50% → 10.5% price relief)

• Demand uplift:

  • Bundling + loyalty + targeted marketing + customization + service: typically 10–20% volume uplift in aggregate (we used ~12% in base)
  • Benchmarks: loyalty/retention studies (5% higher retention meaningfully improves profits), bundling elevates perceived value & repeat purchase

---

7) Worked example: Textile Products

From the chart used in this project:

• Baseline FY2025 (no tariff): $10,812M

• Tariff‑only projection: $4,900M

  • Loss without levers: $5,912M (−54.7% vs baseline)

Levers applied (quantified in model):

• Rebate (RoSCTL ≈ 7%): +$757M
• Cost‑side savings passed to price (≈5.25% of baseline): +$568M
• Demand uplift (bundling + loyalty + marketing + customization ≈ 8.9% of baseline): +$958M

Final projection after levers: $7,183M

• Loss avoided (“business saved”) vs tariff‑only: $2,283M
• % of tariff loss recovered: 38.6%
• Final vs baseline: −33.6% (i.e., at 66.4% of baseline)

---

8) How to reproduce (quick code sketch)

# A) Prepare features
df['year_index'] = df['Year'] - df['Year'].min()
X = pd.get_dummies(df[['year_index','Export Category']], drop_first=True)
y = df['Export Value (USD Million)']

# B) Train XGBRegressor
from xgboost import XGBRegressor
model = XGBRegressor(n_estimators=400, max_depth=4, learning_rate=0.05,
                     subsample=0.9, colsample_bytree=0.9, random_state=42)
model.fit(X, y)

# C) Predict 2025 baseline per category
# ...build X_2025 with year_index+1 and category dummies, predict -> Baseline

# D) Apply tariff shock & levers (Textiles)
T = 0.50; absorb = 0.20; rebate_rate = 0.07
cost_savings_total = 0.21; pass_to_price = 0.50
demand_uplift = 0.12; E = -1.1

net_tariff_to_price = T*(1-absorb)
price_relief = rebate_rate + pass_to_price*cost_savings_total
deltaP_eff = net_tariff_to_price - price_relief
qty_change = E * deltaP_eff

final_value = baseline_textiles * (1 + qty_change) * (1 + demand_uplift)

---

9) Sources & benchmarks (for lever quantification)

• RoSCTL rebate rates (apparel/made‑ups): 6.05–8.2%
• RoDTEP: many lines in chemicals/pharma/iron & steel ~0.5–2%, with several items at ~0.8%
• Logistics optimization potential: 8–15% cost reduction via routing, carrier mix, consolidation
• Vendor rationalization: 3–5% COGS reduction via consolidation & renegotiations
• Commodity forecasting: 5–10% COGS protection via timing and hedging improvements
• Demand levers (bundling/loyalty/marketing/customization/after‑sales): ~10–20% uplift in repeat purchase/volume; retention literature notes strong profit leverage when retention improves

(Adjust ranges per sector.)

---

10) Conclusion (Textile category)

• Tariff‑only would have cut FY2025 textile exports by $5.9B (−54.7%) from the baseline.
• Applying the quantified levers, the model recovers $2.3B of that loss, absorbing ~39% of the shock and landing at $7.18B (i.e., 66.4% of the baseline).
• With stronger absorption (30%), higher pass‑through of savings, or greater demand uplift (e.g., targeted bundling/loyalty), the recovery band improves further—use the sensitivity page to present upside cases to stakeholders.

---