Agent Skills: Time Series Detrending for Macroeconomic Analysis

Time Series Detrending for Macroeconomic Analysis

This skill provides guidance on decomposing economic time series into trend and cyclical components, a fundamental technique in business cycle analysis.

Overview

Economic time series like GDP, consumption, and investment contain both long-term trends and short-term fluctuations (business cycles). Separating these components is essential for:

• Analyzing business cycle correlations
• Comparing volatility across variables
• Identifying leading/lagging indicators

The Hodrick-Prescott (HP) Filter

The HP filter is the most widely used method for detrending macroeconomic data. It decomposes a time series into a trend component and a cyclical component.

Mathematical Foundation

Given a time series $y_t$, the HP filter finds the trend $\tau_t$ that minimizes:

$$\sum_{t=1}^{T}(y_t - \tau_t)^2 + \lambda \sum_{t=2}^{T-1}[(\tau_{t+1} - \tau_t) - (\tau_t - \tau_{t-1})]^2$$

Where:

• First term: Minimizes deviation of data from trend
• Second term: Penalizes changes in the trend's growth rate
• $\lambda$: Smoothing parameter controlling the trade-off

Choosing Lambda (λ)

Critical: The choice of λ depends on data frequency:

| Data Frequency | Recommended λ | Rationale | |----------------|---------------|-----------| | Annual | 100 | Standard for yearly data | | Quarterly | 1600 | Hodrick-Prescott (1997) recommendation | | Monthly | 14400 | Ravn-Uhlig (2002) adjustment |

Common mistake: Using λ=1600 (quarterly default) for annual data produces an overly smooth trend that misses important cyclical dynamics.

Python Implementation

from statsmodels.tsa.filters.hp_filter import hpfilter
import numpy as np

# Apply HP filter
# Returns: (cyclical_component, trend_component)
cycle, trend = hpfilter(data, lamb=100)  # For annual data

# For quarterly data
cycle_q, trend_q = hpfilter(quarterly_data, lamb=1600)

Important: The function parameter is lamb (not lambda, which is a Python keyword).

Log Transformation for Growth Series

Why Use Logs?

For most macroeconomic aggregates (GDP, consumption, investment), you should apply the natural logarithm before filtering:

1. Multiplicative to Additive: Converts percentage changes to log differences
2. Stabilizes Variance: Growth rates become comparable across time
3. Economic Interpretation: Cyclical component represents percentage deviations from trend
4. Standard Practice: Required for business cycle statistics that compare volatilities

import numpy as np

# Apply log transformation BEFORE HP filtering
log_series = np.log(real_series)
cycle, trend = hpfilter(log_series, lamb=100)

# The cycle now represents percentage deviations from trend
# e.g., cycle = 0.02 means 2% above trend

When NOT to Use Logs

• Series that can be negative (net exports, current account)
• Series already expressed as rates or percentages
• Series with zeros

Complete Workflow for Detrending

Step-by-Step Process

1. Load and clean data: Handle missing values, ensure proper time ordering
2. Convert to real terms: Deflate nominal values using appropriate price index
3. Apply log transformation: For positive level variables
4. Apply HP filter: Use appropriate λ for data frequency
5. Analyze cyclical component: Compute correlations, volatilities, etc.

Example: Business Cycle Correlation

import pandas as pd
import numpy as np
from statsmodels.tsa.filters.hp_filter import hpfilter

# Load real (inflation-adjusted) data
real_consumption = pd.Series(...)  # Real consumption expenditure
real_investment = pd.Series(...)   # Real fixed investment

# Log transformation
ln_consumption = np.log(real_consumption)
ln_investment = np.log(real_investment)

# HP filter with λ=100 for annual data
cycle_c, trend_c = hpfilter(ln_consumption, lamb=100)
cycle_i, trend_i = hpfilter(ln_investment, lamb=100)

# Compute correlation of cyclical components
correlation = np.corrcoef(cycle_c, cycle_i)[0, 1]
print(f"Business cycle correlation: {correlation:.4f}")

Dependencies

Ensure these packages are installed:

pip install statsmodels pandas numpy

The HP filter is in statsmodels.tsa.filters.hp_filter.