Hemline Index Theory: empirical analysis with Google data

ABSTRACT

Ability to predict future trends is useful in the fashion industry, as it helps to increase competitiveness, reduce costs and contribute to improvements in the whole supply chain management. Hemline Index Theory (HIT) is a theory that the typical hemline of women’s skirts and dresses (hereafter, referred to as ‘skirts’) can be predicted from economic trends. This research tests the HIT in the Croatian context by using empirical data analysis. Monthly and quarterly women’s skirt data from 2004 to 2019 were collected from the Google Trends data via the Google search volume tool. Once the hemline lengths were calculated, they were processed with the economic growth data. The robust data analysis via the filtered probability of the recession showed that there is little to no evidence that HIT is valid in Croatia. Thus, the state of the economy cannot be used as a good predictor of future skirt length trends.

KEYWORDS:

• Fashion trend forecasting
• Google search volume
• Hemline Index Theory

Acknowledgement

The author is grateful to the two anonymous referees and Simon and Caroline Durk of the ōbex project for language editing support.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 Some of the examples include: economic state and recessions in Bortoli and Combas (2015) and Tkacz (2013); stock market return and volatility in Kim, Lučvijanska, Molnar, and Villa (2019), Škrinjarić and Čižmešija (2019), Škrinjarić (2018, 2019), Bijl, Kringhaug, Molnar, and Sandvik (2016), Habibah, Rajput, and Sadhwani (2017); private consumption in Woo and Owen (2019), Vosen and Schmidt (2011); unemployment in D'Amuri and Marcucci (2017) and Smith (2016); energy consumption in Kaack, Apt, Morgan, and McSharry (2017); tourism spending in Havranek and Zeynalov (2018), Dergiades, Mavragani, and Pan (2018).

2 This is due to the non-stationarity of price series for using it in the regression procedure. Moreover, all variables were used in their original form. The regressions had R² of around 90%, which is an indicator that the variables could be non-stationary and the results could be spurious.

3 This is the most common assumption (Guidolin, 2011a, 2011b). If we assume changing matrix in (4), this would lead to an over-parameterization of the regime switching process (Ang & Bekaert, 2002).