Methodology

This page explains how the research was conducted: where the data comes from, how we built the Economic Anxiety Index, which statistical methods we used, and what the limitations are.

Data Sources

Crypto Adoption Data

On-chain metrics from Glassnode: Bitcoin active addresses, transaction volume, Ethereum addresses, and USDT transfer volume. These are combined into a single Adoption Index that tracks how much the network is being used.

Period: January 2019 – August 2025 (monthly).

Macroeconomic Data

From Federal Reserve Economic Data (FRED): M2 money supply (broad money), inflation expectations (5- and 10-year breakeven rates: T5YIE, T10YIE), yield curve spread (10-year minus 2-year: T10Y2Y), and consumer sentiment. These feed into our Economic Anxiety Index.

Same monthly frequency, aligned with adoption data.

Combined Dataset

All macro and on-chain series are merged into a single monthly panel: CPI (YoY), Fed Funds Rate, U-6 Unemployment, DXY, GDP Growth alongside BTC Active Addresses, BTC Transaction Volume, BTC Holder Score, ETH Active Addresses, and USDT Transfer Volume. Gaps of up to two months are filled by time interpolation, with forward/backward fill used for any remaining missing values.

Period: January 2019 – August 2025 · Frequency: monthly end-of-month · 10 columns · exported as merged_dataset.csv

View or download the merged dataset (CSV)

Base for all models
CPI_YoYFed_Funds_RateU6_UnemploymentDXYGDP_GrowthBTC_Active_Addressesestimated_tx_volumeHolderScoreETH_Active_Addressesusdt_daily_volume

Direct Source Links

  • Glassnode — source for on-chain crypto adoption metrics.
  • FRED — source for macroeconomic indicators used in the Economic Anxiety Index.

Index Construction

The Economic Anxiety Index is built from three components. Each one is z-score normalized (so it has mean 0 and standard deviation 1 over the sample). The final index is the simple average of the three, so each component has equal weight.

Component 1 — Inflation Fear

Based on breakeven inflation rates (T5YIE, T10YIE). Higher expected inflation → higher anxiety.

Component 2 — Monetary Debasement

Based on M2 money supply growth. Faster expansion → higher debasement concern.

Component 3 — Recession Fear

Based on yield curve (T10Y2Y) and consumer sentiment. Inversion or weak sentiment → higher recession fear.

Formula

Economic Anxiety Index = (z(Inflation Fear) + z(Monetary Debasement) + z(Recession Fear)) / 3

where z(·) is z-score normalization over the sample period.

The Crypto Adoption Index is constructed in a similar way. Each crypto activity measure is normalized first, then combined into a single index so that no one component dominates simply because it is measured on a larger scale.

Component 1 — Bitcoin Active Addresses

Measures the number of active Bitcoin addresses. Higher activity suggests broader user participation and network usage.

Component 2 — Estimated Transaction Volume

Captures how much value is moving through the network. Larger transaction volume suggests greater economic use of crypto.

Component 3 — Ethereum Active Addresses

Measures user activity on Ethereum. This helps capture adoption beyond Bitcoin alone.

Component 4 — USDT Transfer Volume

Tracks stablecoin transaction activity. This reflects real usage for transfers, trading, and liquidity within the crypto ecosystem.

Formula

Crypto Adoption Index = (z(BTC Active Addresses) + z(Estimated Transaction Volume) + z(ETH Active Addresses) + z(USDT Transfer Volume)) / 4

Each component is normalized first, then averaged with equal weight.

Statistical Methods

We use standard techniques to measure the link between economic anxiety and adoption, and to assess how well the model generalizes.

  • Pearson correlation — Measures the linear relationship between the Economic Anxiety Index and the Adoption Index (we report r = 0.698).
  • Granger causality tests — Check whether past values of economic anxiety help predict adoption (beyond what adoption’s own past explains). Used to support a directional story, not to prove cause.
  • Time series cross-validation (5-fold) — We split the data into 5 time-ordered blocks, train on earlier blocks, and test on the next. This avoids using future data when evaluating performance and gives a more realistic accuracy (65%).
  • Ridge regression — Used for prediction. Ridge penalizes large coefficients, which reduces overfitting and helps with correlated macro variables (e.g. multiple inflation measures).

Limitations

We report limitations explicitly so the results are interpreted correctly.

  • Small sample size — About 80 monthly observations. More data would strengthen inference and stability of the model.
  • US data only — Macro variables are US-focused. The relationship may differ in other countries or currency regimes.
  • Correlation, not causation — We measure association and predictive ability. We do not prove that economic anxiety causes adoption.
  • 65% accuracy is modest — Better than random (50%) but not suitable for trading or investment decisions. One in three directional calls is wrong.
  • Trained on 2019–2024 — The model has not seen regime shifts or black-swan events outside that period. Unprecedented conditions may break the relationship.