Skip to content

Identifying Bull and Bear Market Regimes with a Robust Rule-Based Method: summary and key takeaways

The paper proposes a bull and bear regime dating rule with one discrete parameter, built by repeatedly filtering peaks and troughs. It compares the rule with two multi-parameter rule-based methods and a Markov-switching model. The new rule matched or beat the others on Sharpe ratio in a simple long-only trading test, most clearly on assets unlike equity indices.

Zegadło, P. · 2022 · Research in International Business and FinanceSummary updated 20 November 2026
Published2022Journal article
Asset focusEquitiesBull/bear regimes
TypeResearchRegimes & Safe Havens
Reading time4 minthis summary

Identifying Bull and Bear Market Regimes with a Robust Rule-Based Method at a glance

The question
Can a rule-based regime classifier with a single, non-optimised parameter date bull and bear markets as well as multi-parameter methods, and work across asset classes without retuning?
Dataset & period
12 daily series from Refinitiv Datastream: 3 stock indices (DAX, DJIA, NIFTY), 3 exchange rates (CHFJPY, EURPLN, GBPUSD) and 6 commodities (Copper, Corn, Gold, Oil, Palladium, Sugar). Each series runs from the start of its available history to the end of 2020. Observation counts range from 5738 (EURPLN) to 17,772 (DJIA).
Method
The rule takes daily closes, marks local peaks and troughs, then filters those extrema again several times, retaining a peak only if it exceeds the peaks on either side, and a trough only if it is below the troughs on either side. Peaks and troughs must alternate. Four filtering passes are used for daily data, and the paper suggests fewer for weekly data and more for hourly data. The out-of-sample test goes long when a bull regime is detected and exits in a bear regime, entering one day after detection, with round-trip costs. Sharpe ratio differences are tested with White's Bootstrap Reality Check.

Key findings of Identifying Bull and Bear Market Regimes with a Robust Rule-Based Method

  • The rule has a single parameter, the count of filtering passes, fixed at 4 for daily data without optimisation.
  • With typical settings, the new rule and the Pagan-Sossounov method agree on 91.3% of days on average. Agreement with the Lunde-Timmermann method is lower, at 86.2%.
  • Four passes on daily data produce regimes that typically last from 50 to 500 days.
  • Across most series, bull phases run longer and swing further than bear phases. Gold, Sugar and EURPLN break this pattern.
  • Bear regimes tend to show higher variance and larger absolute daily returns for most assets, consistent with the leverage effect.
  • In the long-only trading test, the new rule ranked first in more cases than any other and never significantly underperformed. On GBPUSD it was the only rule with a positive Sharpe ratio.

Robustness, as reported in the paper

  • Out-of-sample trading tests use a one-day delay after detection, so the identification lag of rule-based methods is fully counted.
  • Round-trip transaction costs of 0.1% are charged, which penalises methods that switch regime often.
  • The Markov-switching benchmark is re-estimated daily to mimic a real investor. It opens very many positions and suffers from costs.
  • Significance is judged at the 10% level using a stationary bootstrap version of White's Reality Check, with full p-value tables in the appendix.

Caveats and limits

  • The trading rule is a goodness-of-fit device, not a full strategy. The paper says no rule consistently and significantly dominates buy-and-hold.
  • All rule-based methods detect regimes with a delay, which limits real-time use. The paper lists reducing this delay as future work.
  • In fast-oscillating prices the rule can flag regimes with trivially small amplitude. A neutral state needs an added amplitude threshold, which brings back a subjective parameter.
  • Results are for one sample ending in 2020, and the rule's own settings for non-daily data are suggestions rather than tested values (inferred).

Items marked (inferred) are our reading of the design, not statements by the authors.

Why it matters

Regime labels from a method with no tuned thresholds are easier to compare across equities, currencies and commodities. That helps with cross-asset studies of correlations in bull and bear states. The paper also notes the rule could serve as a filter for trading signals, though it stresses more work is needed.

Questions about this paper

How many parameters does the new method use?

One: the count of filtering passes, fixed at 4 for daily data. It is not optimised and scales with data frequency.

How different are the regimes from the older methods?

On average the new rule matches the Pagan-Sossounov classification on 91.3% of days and the Lunde-Timmermann one on 86.2%.

Did the rule win the trading comparison everywhere?

It ranked first in the most cases and never significantly underperformed another rule. On several assets the differences were not significant at the 10% level.

Where can I read the original paper?

The original is at its publisher: https://doi.org/10.1016/j.ribaf.2021.101603. We link to the paper and do not host the PDF.

Follow new quant research

Download MacroFinance market data as CSV or Excel, or pull it by API, and test the ideas in these papers yourself. New summaries are added as we review them.

Create a free account
Summaries are written by MacroFinance from the published papers, in our own words, for information only. They are not investment advice and may omit detail. Papers are not hosted here: each page links to the original at its publisher.