Rally episodes in Meme coin perpetual futures : a rule-based detection and derivative market characterization
Tongwa, Tatiana Nanette (2026)
Pro gradu -tutkielma
2026
School of Business and Management, Kauppatieteet
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi-fe20260708110050
https://urn.fi/URN:NBN:fi-fe20260708110050
Tiivistelmä
Meme coins are cryptocurrency assets that are greatly influenced by social media attention. These coins experience price rallies that are short and sharp which are later characterised by heavy losses incurred by traders and negative spillover effects on major cryptocurrencies such as Bitcoin. As a result, this thesis works towards building a transparent rule-based method strictly using price and time, that identifies rally episodes across a set of meme coins and splits each identified rally episode into four phases namely: pre-rally, rise, fall, and post-rally. The research work then proceeds to describing and predicting these phases with the help of the derivative market variables including open interest, funding rate, long/short ratio, and traded volume using a multinomial logistic regression, leaving out price from the model design.
The meme coin set and daily data used for this analysis are obtained from CoinGecko and Coinalyze respectively, over a time frame from April 2022 to April 2026. The rally detection rule follows both the Bry-Boschan and Pagan-Sossounov methodology and a single-exchange constraint for data collection. With the use of a multinomial logistic regression, the classification of the rally episodes phases is done using two models: a same-day descriptive model and a lagged predictive model, the predictive model estimated with inverse-frequency class weighting and walk-forward cross-validation. The main evaluation metric is the macro-averaged F1 score.
The rule correctly identifies 76.9% of a curate set of famous documented rallies in a sample of 363 rally episodes from 62 coins. The derivative variables define the phases together with each other. They have a consistent buildup of leverage, peak and unwind cycle and their main marker is volume. The derivatives also provide a better view on extremes of funding pressures and positioning when the rally is in a fall phase rather than at its peak. The lagged variables include real next-day information that has improved over two times of the descriptive model baseline F1 macro level (0.275 to 0.122). Hence, this thesis offers a four phase detection framework for rally life cycles using a derivative-only approach as well as an empirical account of the rally life cycle, and finally a derivative-only method for detecting speculative episodes within markets that expose retail investors to real levels of risk.
The meme coin set and daily data used for this analysis are obtained from CoinGecko and Coinalyze respectively, over a time frame from April 2022 to April 2026. The rally detection rule follows both the Bry-Boschan and Pagan-Sossounov methodology and a single-exchange constraint for data collection. With the use of a multinomial logistic regression, the classification of the rally episodes phases is done using two models: a same-day descriptive model and a lagged predictive model, the predictive model estimated with inverse-frequency class weighting and walk-forward cross-validation. The main evaluation metric is the macro-averaged F1 score.
The rule correctly identifies 76.9% of a curate set of famous documented rallies in a sample of 363 rally episodes from 62 coins. The derivative variables define the phases together with each other. They have a consistent buildup of leverage, peak and unwind cycle and their main marker is volume. The derivatives also provide a better view on extremes of funding pressures and positioning when the rally is in a fall phase rather than at its peak. The lagged variables include real next-day information that has improved over two times of the descriptive model baseline F1 macro level (0.275 to 0.122). Hence, this thesis offers a four phase detection framework for rally life cycles using a derivative-only approach as well as an empirical account of the rally life cycle, and finally a derivative-only method for detecting speculative episodes within markets that expose retail investors to real levels of risk.