1. Introduction

This report shows the bubble methodology applied to songs. Just like for the real estate market, this report applies the bubbles method for detection of explosive behaviors to classify a song as a hit or not and discover in which date it started to be a hit.

2. Data Used

To do so, the dataset used is composed by streams extracted from Spotify database. The dataset contains a daily and a weekly stream for each one of the first 200 songs. In this report we are gonna analyze those songs that have been among the top 10 at some point.

3. BADF Sequence

The Backwards ADF Sequence displays the statistics of the Augmented Dickey-Fuller test for different periods of time. As reference a 95% quantile is used. Every time the series presents a explosive behavior the BADF Sequence exceeds the 95% quantile. The graphs only captures the positive explosive behaviors because that is the interest of this report. When a song presents a fast decline, the BADF Sequence adjusts the test with a time trend, making the sequence smoother. The plots below show the behavior for the songs that have been among the top 10 at some point.