Nestlé Commodity vs Stock Performance

End-to-end data pipeline and statistical analysis examining whether Nestlé's key commodity input prices have a measurable relationship with Nestlé's stock price (NESN.SW on the Swiss Exchange).

427 months of data · Jan 1990 – Feb 2026 · 8 commodities · 2 data sources · OLS regression

---

Overview

Nestlé's cost base depends heavily on raw commodities — coffee, cocoa, sugar, wheat, corn, palm oil, soybeans. The question this project explores: do movements in these commodity prices actually move Nestlé's stock?

The pipeline covers three phases:

1. Data Engineering — two separate pipelines for sourcing commodity prices (World Bank Excel + Alpha Vantage API), cleaning, unit conversion, and building a star schema ready for Power BI
2. Exploratory Analysis — correlation heatmap, lagged correlation (0–6 months), inter-commodity clustering
3. Statistical Modelling — OLS regression identifying which commodities explain stock return variation

---

Key Findings

• Cocoa is the only commodity with a meaningful negative correlation with Nestlé's stock returns (r = −0.16). The market reaction is immediate — same month, no lag
• Coffee shows a slight positive signal at lag_1m — rising coffee prices likely signal strong global demand, which benefits Nespresso and Nescafé revenue
• R² = 3.8% — two commodities explain ~4% of monthly stock return variation. The remaining 96% is driven by macro conditions, FX (CHF/EUR), earnings surprises, and broader market sentiment
• Nestlé's diversification absorbs most commodity shocks — 400+ brands across categories prevent any single input from dominating financial performance

---

Notebooks

01_World_Bank_Pipeline.ipynb

Full historical pipeline using the World Bank Pink Sheet — the gold standard for commodity data, covering 71 commodities from 1960.

• Cleans non-standard Excel layout (metadata rows, custom date format)
• Filters to 8 Nestlé-relevant commodities
• Reshapes wide → long format via pd.melt()
• Engineers features: YoY % change, 12-month rolling average
• Builds star schema (dim_commodity, fact_prices) exported as CSV for Power BI
• Downloads Nestlé stock price via yfinance, merges with commodity data
• Runs correlation analysis, lagged correlation, and OLS regression

02_Alpha_Vantage_Pipeline.ipynb

Companion pipeline using the Alpha Vantage REST API — designed to top up recent months when the Pink Sheet hasn't been updated yet.

• Reusable fetch_commodity() function with built-in error handling and rate limit management
• Converts raw API units (cents/lb) to $/kg and $/mt to match World Bank schema
• Handles Alpha Vantage quirks: . as null placeholder, newest-first sort order
• Visualises indexed price trends — all commodities normalised to 100 for fair comparison

Recommended architecture:

World Bank (1960 → last month)  +  Alpha Vantage (last 1–2 months top-up)
                        ↓
              merge on (date, commodity)

---

Commodities Covered

Commodity	Category	Unit	Pipeline
Coffee Arabica	Beverages	$/kg	Both
Coffee Robusta	Beverages	$/kg	World Bank
Cocoa	Confectionery	$/kg	World Bank
Sugar (World)	Sweeteners	$/kg	Both
Wheat (US SRW)	Grains	$/mt	Both
Corn (Maize)	Grains	$/mt	Both
Palm Oil	Oils & Fats	$/mt	World Bank
Soybeans	Oils & Fats	$/mt	World Bank

---

Tech Stack

Python 3.9
├── pandas          # data wrangling, reshaping, feature engineering
├── yfinance        # Nestlé stock price download
├── requests        # Alpha Vantage REST API
├── scikit-learn    # linear regression
├── statsmodels     # OLS with significance testing
├── seaborn / matplotlib  # static visualisation
└── plotly          # interactive indexed price chart

---

Project Structure

nestle-commodity-analysis/
│
├── notebooks/
│   ├── 01_World_Bank_Pipeline.ipynb
│   └── 02_Alpha_Vantage_Pipeline.ipynb
│
├── data/
│   ├── dim_commodity.csv       # dimension table, 8 rows
│   └── fact_prices.csv         # fact table, 6,117 rows
│
└── README.md

---

What I Learned

Data Engineering

• Multi-key lookup — matching on commodity AND date simultaneously; why a single-column join fails here
• Sort order matters for indexing — iloc[0] silently produces wrong results on descending data
• Unit conversion from raw API — cents/lb → $/kg is one line of code but requires reading API metadata carefully
• Star schema design — structuring data for Power BI consumption from the start saves rework later

Statistics

• Returns vs prices — raw prices create spurious correlation due to shared upward trend; pct_change() isolates genuine co-movement
• Lagged correlation — testing 0–6 month lags reveals whether markets react immediately or with delay
• Honest R² interpretation — 3.8% is not a failure; it reflects the reality of a diversified company

Domain

• Diversified FMCG companies are naturally hedged — no single commodity dominates the P&L
• Cost-side shocks (cocoa) hit stock price immediately; demand-side signals (coffee) lag by one month

---

Limitations

• Yahoo Finance NESN.SW data starts from ~1990, limiting the analysis window despite commodity data going back to 1960
• Regression uses OLS without controlling for autocorrelation in residuals — statsmodels full summary recommended before drawing strong conclusions
• Alpha Vantage free tier excludes Cocoa, Palm Oil, and Soybeans

Next Steps

• Add EUR/CHF FX rate and market index (S&P 500 or MSCI World) as control variables
• Test on quarterly frequency — may produce stronger signal as earnings reflect actual commodity costs
• Extend to peer comparison: Unilever (ULVR.L) or Mondelez (MDLZ)
• Build Power BI dashboard using exported CSV files

---

Author

Roman Honta · Junior BI Analyst at Nestlé Ukraine
LinkedIn · Portfolio