20180928-法国巴黎银行-Inferring_credit_cycles_from_EM_asset_prices_15页_1mb

Summary of "Inferring credit cycles from EM asset prices"

Core Content

This document presents a novel approach to inferring the phase of economic cycles in emerging markets (EM) using only asset prices, rather than traditional macroeconomic variables. The methodology was developed by BNP Paribas Brasil S.A. and tested on Brazil, Mexico, and South Africa.

Main Viewpoints

• Alternative Approach to Economic Cycles: The document challenges the conventional method of using macroeconomic variables to assess economic cycles. Instead, it proposes a model that uses financial market data to infer the current phase of the economic cycle.
• Model Design: The model uses a combination of financial indicators such as interest rates, equity, bonds, break-evens, and FX performance. These are normalized and smoothed using a Hodrick-Prescott (HP) filter to isolate trends and reduce noise.
• Sinusoidal Fit: A sinusoidal function is fitted to the smoothed data to identify the four stages of the economic cycle: early recovery, mid recovery, late recovery, and recession.
• Adaptive and Dynamic Model: The model is adaptive, using a rolling window approach to ensure that it can respond to changes in the economic cycle over time.
• Phase and Duration Estimation: The model not only identifies the current phase of the cycle but also estimates the duration, speed of acceleration/deceleration, and the expected remaining time of the cycle.
• Strategy Recommendations: The model results are used to inform investment strategies based on the difference between the perceived cycle phase from market prices and the macroeconomic view.

Key Information

Brazil

• Current Phase: Early recovery phase.
• Peak Prediction: Expected to reach the peak in approximately 16 months.
• Index Behavior: The model's index is close to neutrality, indicating a balanced economic perception.
• Historical Context: The model identified the 2016 Temer tantrum as a long recession due to rapid changes in the economic environment.

Mexico

• Current Phase: Still in the early stages of recovery.
• Peak Prediction: Expected to reach the peak in approximately 16 months.
• Index Behavior: The index suggests a similar cycle to Brazil, but with a slower pace of recovery.
• Historical Context: The economic cycle was less defined, with recovery and slowdowns between 2011 and 2015.

South Africa

• Current Phase: Close to the peak of the cycle, with a prediction that the peak may be reached in the next three months.
• Index Behavior: The model identified a sequence of cycles with different durations, including a deep recession in 2016.
• Historical Context: The model showed a long mid-recovery phase before 2008, leading to a higher average duration for that cycle.

Methodology Overview

1. Normalization: A 30-month rolling window was used to normalize the financial data to a scale between -1 and +1.
2. HP Filter: Applied to smooth the data and isolate the trend component.
3. Sinusoidal Fit: Used to model the cycle with four phases, defined by the angle of the sine function.
4. Constraints: The model was constrained to prevent abrupt changes in the angle, limiting it to a maximum change of 60° upward and 20° downward per observation.
5. Cycle Estimation: The model estimates the historical duration of each cycle phase and the expected remaining time based on the current speed relative to historical averages.

Implications

• The model provides a leading indicator for economic activity in EM countries.
• It helps in identifying the current phase of the cycle and can support investment decisions.
• The results can be used to recommend strategies such as bear flattening to bull steepening, and equities outperforming fixed income during recovery phases.

Technical Appendix Highlights

• Normalization Formula: Used to scale data between -1 and +1.
• HP Filter Formula: Used to smooth the time series data.
• Sinusoidal Fit: Defined by the formula $ y_{\text{partial model}} = i_0 + a \cdot \sin \left( \frac{2 \pi}{T} \cdot t + \varphi \right) $, where $ T $ is the cycle period, $ t $ is the time point, and $ \varphi $ is the phase.
• Rolling Window: The model uses a rolling window of $ T / 4 $ to adapt to changes in the cycle.
• Final Model: The final model is a combination of partial models, achieving an R-squared of 99.7%.

Legal Notice

• The document is a marketing communication and not investment research.
• It is intended for professional clients and eligible counterparties.
• The information is based on public sources and may not be independently verified.
• The document may contain back-tested performance data, which is for illustrative purposes only.
• BNP Paribas may have conflicts of interest and may engage in transactions inconsistent with the views expressed in the document.