Behavioural heterogeneity in wine investments

ABSTRACT

We introduce a heterogeneous agent model to explain the dynamics of fine wine investments. Our results show evidence of the existence of both fundamentalists – those who trade on mean-reversion towards a fair value – and chartists – those who extrapolate recently observed price trends – in the wine market. Moreover, we document that market participants switch between the two trading strategies, allocating more weight to the strategy that has been the most accurate in forecasting wine index values in the recent past. This switching behaviour can explain the large variations in index values (bubbles and crashes) that are observed in the fine wine market.

KEYWORDS:

• Wine investments
• Heterogeneous agent models
• behavioural finance

JEL CLASSIFICATION:

• C22
• G11

Acknowlegment

We would like to thank the editor and the referees for their useful suggestions. We would also like to thank the participants at the 2017 Annual AAWE Meetings (Padova, Italy) for their useful comments and suggestions. We would like to thank Liv-Ex for providing us with regional index and constituent data.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

¹ According to Wine Spectator sales at major wine auctions increased from 90 million USD in 2002 to 381 million USD in 2017.

² This low correlation with financial assets should be put in perspective as investing in fine wine is not as liquid as investing in traditional financial assets (Masset and Weisskopf 2018b).

³ Dimson, Rousseau, and Spaenjers (2015) document a return of 10.9% p.a. over the period 1900–2012, compared with a 6.1% return p.a. on US government bonds.

⁴ London International Vintners Exchange (www.liv-ex.com).

⁵ See for instance http://www.winespectator.com/webfeature/show/id/46287. However, this bubble-like behavior has been recognized earlier. For instance, Robinson (1998) notes that: ‘What is more extraordinary is the wild price variation at the very top end -fine wines. Demand bubbles up mysteriously, apparently fuelled by fashion and rumour as much as by intrinsic quality’. We further note that bubble-like behaviour is also observed in other markets for emotional assets, such as the art market (see e.g. Kräussl, Lehnert, and Martelin 2016 or Penasse and Renneboog 2017).

⁶ Although it is not the purpose of this paper to assess bubbles in the context of Phillips, Wu, and Yu (2011), (2015)), in unreported tests, we find that there are indeed episodes with explosive behaviour in our wine index. These results are available on request.

⁷ We thank the referee for pointing this out to us.

⁸ Note that throughout the paper we use the terms ‘index value’ and ‘price’ interchangeably. Where we use the term ‘price’ in the context of our analysis, we essentially refer to index values.

⁹ This switching behavior can be interpreted in two ways. Either investors change their beliefs and instead of following one strategy decide to follow another strategy, or there is heterogeneity among investors, and investors with different beliefs (e.g. investors, consumers and collectors) choose to trade at different times when they realize that their trading strategy is more accurate.

¹⁰ See Oczkowski and Doucouliagos (2014) for a recent survey on wine price determinants.

¹¹ Wine can be considered a prime example of an experience good as a consumer can only elicit its true intrinsic quality once purchased and consumed.

¹² Other empirical frameworks that consider fundamentalists-chartists have been considered in the literature, such as Westerhoff and Reitz (2003) and Menkhoff, Rebitzky, and Schröder (2009) who look at the impact of fundamentalist/chartist beliefs on the dynamics of exchange rates.

¹³ Chen, Chang, and Du (2012) provide an excellent overview on the empirical HAM literature.

¹⁴ Several competing approaches used to detect bubble-like behaviour have been employed on tangible or alternative asset classes. A common method is the superior-ADF (SADF) test of Phillips, Wu, and Yu (2011). For instance, Gilbert (2010) applies the SADF procedure to three CBOT future prices, wheat, corn and soybeans, concluding that during the commodity boom and bust of 2007–2008 there was explosive behaviour in soybean future prices. Phillips et al.’s SADF test is also used by Etienne, Irwin, and Garcia (2014). They find that there are speculative bubbles in 12 agricultural markets. More in line with our study, Czupryna and Oleksy (2015) find explosive behaviour in the Liv-ex 50 index. Figuerola-Ferretti, Gilbert, and Mccrorie (2015) use a multi-bubble generalization of the SADF test, GSADF, proposed by Phillips, Shi, and Yu (2015) to examine the price behaviour of the six main London Metal Exchange non-ferrous metals prices. They detect periods of mild explosivity in copper, nickel, lead, zinc and tin, but not in aluminium. Similarly, Tsvetanov, Coakley, and Kellard (2016) apply the GSADF test to crude oil prices, finding significant bubble periods. Alternative bubble detection methodologies have also been used, such as Zhou and Sornette’s (2009) D-test (oil price in Zhang and Yao 2016); van Norden and Schaller’s (1993) switching regression model (grains, softs, animals and woods, precious metals, and energy in Brooks, Prokopczuk, and Wu 2015); or the momentum threshold autoregressive (MTAR) approach (US corn, soybean and wheat prices) of Adämmer and Bohl (2015).

¹⁵ Of course, collectors and consumers may also behave as speculators, or their demand can be random as captured by the noise term in Equation (7).

¹⁶ Alternatively, there may be heterogeneity between investors, and specific types of investors may trade at different points in time depending on how well their strategy performs.

¹⁷ Note that in the empirical estimation we use logs of index values and so R_t is the continuously compounded return.

¹⁸ Equations (4), (5) and (7) can be combined into a single (non-linear) equation and estimated directly using quasi-maximum likelihood. This is possible as Equation (7) is a non-linear polynomial of $R_{t+1}$, with the fundamental price $V_t$ given as an exogenous variable (see Kouwenberg and Zwinkels 2014).

¹⁹ Portfolio constituents as at the end of our sample period are shown in Appendix A.

²⁰ We consider a range of alternative definitions of the fundamental value. We present the results for these alternative definitions in Section 5.3. Our main results are not altered by these alternative definitions.

²¹ These are the vintages that are included in the index (see Appendix A), which, to a varying degree coincide with good vintages.

²² Although a comparison between competing bubble-detection approaches is beyond the scope of this paper, we have applied the generalized version of SADF test, GSADF, of Phillips, Shi, and Yu (2015) which can detect the presence of multiple bubbles over the Liv-Ex Fine Wine Investables Index. The unreported results confirm the presence of explosive behaviour in wine index, detecting two significant bubbles, from June 1994 to November 1997, and from October 2005 to September 2007. These two bubbles coincide with an average fundamentalist weight of 23.12% which represents a strong presence of chartists, confirming the main results of our paper.

²³ We observe that the switching parameter is insignificant in one case. It is important to points out that the t-statistic for the switching parameter is not well-defined as the parameter enters the model in non-linearly (Teräsvirta 1994). Boswijk, Hommes, and Manzan (2007) also report an insignificant switching parameter. However, the Likelihood Ratio test is conclusive in assessing whether the switching model is better than the static model, and in the case of that particular insignificant switching parameter, the LR statistic is significant.