1. Introduction
Is wine a good alternative investment? Can this investment have interesting returns? Not always a theme of agreement and concord between investors, emotional assets besides the normal emotional and collectable understandable appeal can be seen as an alternative to the regular markets and as an escape to the volatility of these or simply as a diversification of the investment Portfolios. Several studies have been performed over the years regarding this kind of assets and its possible returns with a wide broadband of results. This kind of investments has had a particular great increase of interest in the eyes of investors as they have shown along years that they can deliver great and stable returns comparing to other kind of investments including government bonds, treasury bills, and gold over the long run with very low volatility (Dimson and Spaenjers, 2014).
Investing in art, wine, stamps, furniture, watches, classic cars or other collections can at first glance be strange and seen as a passion only with the sole purpose of serving of pleasure and ownership pride to the collectors according to Dimson, Rousseau and Spaenjers (2015), however if we can put aside the emotional part of such assets and focus on the pure financial result of these investments, by taking into account the returns given by these assets and concluded in studies along the time it becomes quite clear why many investors look at them as an opportunity as they seem to provide great and stable returns in the long term thus being a very good option to regular financial investments.
Our particular interest and the main focus of this study is in wine investment as this is a tendency in the last years especially among wealthy investors, as confirmed by a Barclays report from 2012 where the conducted survey concludes that wealthy individuals allocate as much as 2% of their wealth to wine investment.
Recently we also observe a great increase in companies specialized to the wine investment business as well as the creation of vehicles to these investments as funds, bonds and indexes based on the value of wine, such as Live-ex Fine Wine 50 or Live-ex Bordeaux 500 and the Wine Investment Fund, however this kind of investment mainly focus on great and very well-known wines as they are worldwide recognized, considered to be the most exclusive and therefore the most valuable for the purpose.
The interest in Portuguese wines is increasing and in a general and international opinion from experts Portugal is recognized as to be the place of creation of some of the most amazing wines according to recent specialized critics , especially when it becomes to fortified wines with the well-known Port Wine, produced in the Douro valley.
Due to the well-known aging potential of this kind of wine, which is a lot more extendable than most wines, and the great value given by experts and collectors we believe that by filling all the necessary characteristics, Port can be a good alternative to investment and can deliver good returns on its storage if done solely with investment purposes.
The main objective of this study is to understand if Port wine can be a good investment through the creation of an index which represents the evolution of Port and calculation of returns based on the trades made along a certain period of time for a particular set of combinations of Port brands and Vintage years.
As very few studies of this kind related to Port wine have been made so far, we believe the subject will be of great interest and a good increase to the current and already existent literature on the subject and, at the same time, it will also allow us to understand if Port wine is able to compete with other wines in terms of returns and therefore can also be taken in consideration for investment purposes.
This dissertation is structured as follows: in chapter 2, a literature review of the topic is made, giving an overview over similar studies as well as analyzing the insight. Chapter 3 is composed by a view on the wine investment aspect, justifying the interest that Port wine might have to these investments. In chapter 4 we present the data for our analysis followed by the explanation for the methodology used. To finalize we present our analysis in chapter 6 and our conclusions in chapter 7.
3. Why Port and why Vintage?
Port is a very well-known Portuguese wine with several centuries of history, the region where it is produced is considered the oldest officially demarcated wine region in the world and its wine has been traded for centuries all over the world.
Due to its organoleptic characteristics, Port wine is capable of improving in quality over time (aging), progressively conferring higher prices on the product as it is stated by Correia, Rebelo and Caldas (2015).
This is one of the reasons why we believe Port can be considered as an option when it comes to long term investment decisions, as these wines normally have a great capacity of aging. This is especially true on a specific Port category named Vintage Port, which is considered to be the pinnacle in quality when it comes to Port wine as only the best grapes of the best properties in the Douro valley are used for the lots, which are used to the production of these wines, adding to this selection the vintage is only declared when the harvest is considered exceptional, which usually only happens 3 or 4 years in a decade (Mayson, 2016), as all the Port brands like to keep their high standards when it comes to this category of Port wine in order not to disappoint the consumer.
Vintage Port is very simple to produce, they are wines from a single year bottled without any treatment or filtration which spend a maximum of 2 years aging in bulk (Mayson, 2016). After bottled these are known to maintain its fundamental characteristics and improve with long aging.
This special category designated as Vintage Port only accounts for a tiny fraction of total production, however it has a great impact on brand reputation and of course a big financial impact on producers.
Vintage Ports are frequently part of auctions worldwide as they are a very desired collectable and investment asset for the interested public all over the world.
For all these characteristics we believe that Vintage Port can be compared to the highest quality wines of 1st choice used for this kind of investment which are the ones with highest returns according to Masset and Henderson (2010).
4. Data
In this chapter we present our data source and the method used for data collection. We also explain how we achieved the final database to perform our analysis.
4.1. Data source
In order to be in line with past studies we used the same approach as most of the existing literature and we collected data from past auctions as these are the main vehicle for trades such as wine, transforming them into the most important platform for data regarding this type of transactions.
By using this type of data we also prevent any kind of deviation that could come from retail prices that might incorporate other factors allowing us to focus only on the actual value that the specific market gives to these assets.
For our approach we followed the method used by Kourtis, Markellos and Psychoyios (2012) and used the same source to collect our data for the study. The data was obtained from a very relevant online database identified as WinePrices.com, which gathers information from several specialized auction houses worldwide on past auctions, which represent together the majority of the current wine auction market. The observations are identified with lot numbers for better identification.
The database WinePrices.com is a free online resource which provides wine price information from auctions and retail worldwide, this aspect allowed us to have data from several locations worldwide and therefore minimize the risk of our study being affected by possible local sales phenomena which might cause deviations to our study. All data is freely available and can be consulted in https://www.vinfolio.com/do/wineprices/home.
Included in the data provided we can find results from auctions at Christie’s which is one of the most important auctioneers in the world, which has included wines in their auctions since their first auction in 1766 in London. This long tradition in presenting wines in their lots turns Christie's in a very good source of such information as wine prices (Dimson, 2015). Prices are gathered from the Christie’s locations in New York, London, Los Angeles, Paris, Amsterdam, Bordeaux, Chicago, Hong Kong, Germany, Geneva, South Kensington and Glasgow.
Adding to the data from Christie’s we also have other well-known auction houses such as Acker, Merrill & Condit in New York and Hong Kong, Sothebys in New York, London and Hong Kong, Zachys in New York, Los Angeles, Las Vegas and Hong Kong, Bonhams in London and Hong Kong, Bonhams & Butterfields in San Francisco, Morrells in New York, Hart, Davis, Hart in Chicago, Edward Roberts International in Chicago and San Francisco, Bloomsbury/Sokolin in New York, WineGavel in San Francisco and Spectrum Wine Auctions in Dana Point and Los Angeles.
This great variety of locations allows us to have extended data worldwide and due to this great variety we could also have some issues with different currencies however all sales prices registered in WinePrices.com in different currencies than U.S. dollars are converted to Dollars as of the date of the sale, which can be found in the data. This option from the data provider facilitates our collection of information and creates a standard for the exchange rate used.
4.2. Data collection
To create our dataset we conducted our data collection based in all the main Port wine brands detained by the companies associated of “AEVP – Associação das Empresas de Vinho do Portoâ€, which is considered one of the main association between Port traders with their associates representing 90% of all Port trades , giving it a very important position in the Port wine market. The list of associated companies is public and can be consulted in the association website http://www.aevp.pt/ASSOCIADOS.
For our analysis we take into account one of the Port wine types known as vintage Port, more specifically each brand classic vintages, as these are known to be the best and rarest Port wines being only declared in special years where the wine is of excellent quality and is up to the highest standards and it is also known for its capacity of ageing which transforms it in one of the most valuables and appreciated Ports worldwide and therefore is considered an asset that gains value through time.
The focus for our data collection is therefore specifically in single Vintage Port bottles in good condition from standard size 0,75L as these are the most common and easily bought for investment purposes.
From the research done we were also able to understand that some of the trades are done in bulk, normally boxes of 6 or 12 bottles as this is the most commonly way of trading wine which is already intended to serve investment or collection purposes, however in these cases of uniform lots of wine ( with same brand, year and size ) the data provider already supplies us with the price per bottle realized in U.S. dollars inclusive of the buyer's premium for the auction house and location at the time but exclusive of sales taxes or VAT.
Using our data source Wineprices.com and conducting a research for classic vintage of the 35 brands we were able to obtain a total of 14960 observations for auction hammer prices, including brand, vintage year, selling price and selling year. The gathered initial set of data is therefore bigger than previous studies focused on Port wine.
Our original data contains observations from auction sales occurred from 1988 to 2018 for 24 different brands as only these from the original 35 researched have auction data and a total of 68 different vintage years ranging from 1815 to 2012.
This translates into a count of 346 different combinations of brand and vintage year following the same approach of Dimson, Rousseau and Spaenjers (2015), some with several sales during the period analyzed.
4.3. Final database
To construct our final database, for all combinations which have more than two sales during the observations period and according to Shiller (1991) we treat them as separate pairs without creating overlap in the holding period, meaning all are treated consecutively and without repetition.
To be in accordance with the methodology used we also drop all single sale observations, meaning all wines sold only once during the period are not part of the final observations, same applies to sales without a pair, which are in fact single sales.
Also for all observations in the same year of the same combination of brand and vintage year and as our objective is to create an annual index, instead of dropping any of the observations we decided to follow a similar approach as Dimson, Rousseau and Spaenjers (2015) and we average the price of the observation which take place in the same year for the same combinations.
After all the necessary adjustments for our analysis we count on 2727 individual annual observations to create 2434 pairs of sales of 293 combinations with sales dates between 2000 and 2018, with a price range between 4,01 and 9108 USD.
The 2727 observations of sales will from now on be our base data including data from 24 brands and 52 vintage years from 1847 to 2011.
5. Methodology
This chapter presents the explanation for the methodology used for our analysis.
In line with our research and according to Table 1 we decided to apply the repeated sales method to our data, as it is the most commonly used in similar studies.
This method was initially proposed by Bailey, Muth and Nourse (1963) to construct real estate price indexes and has since then been applied to other infrequently traded assets (Lucey and Devine, 2015) being considered suitable for the type of data which is normally available for wine trades.
The original method uses data on properties sold more than once during a certain period and the necessary information to estimate the model is reduced as it consists only in price, sales date and identification of the property. Comparing to the real estate market we have a bigger number of sales of bottles of the same wine facilitating the estimation of the trend of the wine market (Lucey and Devine, 2015).
According to Shiller (1991),"the method estimates an index of log prices by regressing log price changes on a matrix of dummy variables".
The regression estimates an index by considering the repeated sales of the same item. In our specific case as we are analyzing wine this means to match pairs of combinations of brand and vintage year sold in different dates to obtain our index.
Shiller (1991) explains the model as following: “The matrix of independent variables is the n x T matrix Z whose ijth element is -1 if the first sale of house i occurred in period j, is 1 if the second sale of house i occurred in period j, and is 0 otherwise. The first column of Z corresponds to t = 1, there is no column for t = 0 since the estimated (log) index will be zero at t = 0 (the base year) by construction, so that its antilog will be 1 at t = 0. The dependent variable vector y has ith element equal to the change in log price for the ith house, using pij = ln(Pij), where Pij is the price of the ith house at time j. The model to be estimated asserts that y = Zγ +℮, where the ith element of γ is the log price index for time t, and for the purpose of computing standard errors it is assumed that the elements of the vector of error terms ℮ are independent of each other, reflecting the notion that individual house price variations unrelated to the city-wide variations are due to idiosyncratic value changes. Then the estimated log price index for time t is the tth element of the ordinary least-squares regression coefficient vector γ ̂ = (Z^' Z)^(-1) Z^' γ.
If the change in log price of a house is given by the change in a true city-wide price index γ plus a zero-mean error term that is uncorrelated with the error terms associated other houses, and if the variance of this error term is the same for all houses, then the standard error matrix of γ Ì‚ has the usual form s^2 (Z^' Z)^(-1).â€
Using the same example as Shiller (1991) for the estimator we consider a very small dataset of five houses and three periods, to estimate two index values. If houses 1 and 2 are sold in periods 1 and 2, houses 3 and 5 are sold in periods 0 and 1, and house 4 is sold in periods 0 and 2, we have the following:
The normal equations Z'Zγ ̂ = Z^' y interpretation is that the ith equation results in the estimated log index for the ith period which is the average log price of all assets sold in that period minus the average of their base-period log price inferred from their other sale price using the estimated index.
For the example the model normal equations are:
The equation for γ ̂_1 (2) which is the index for the first period, results from the four houses sold in that period, where houses 3 and 5 have another sale in the base period, and houses 1 and 2 are sold in period 2. The equation is corrected by subtracting γ ̂_2 to infer a base-period price. The same way the equation γ ̂_2 (3), results from the three houses that were sold in period 2 minus the average inferred log price of these three houses in period 0.
The estimated log price index is based on averages of log price changes of each asset, if we take exp(γ ̂) as as our index level, it is based on the geometric average of prices.
6. Analysis
6.1. Descriptive statistics
Table 2 below shows the descriptive statistics for the observations in our final database. All data was collected from the online auction database wineprices.com as described above in chapter 4. Panel A shows the number of yearly observations (after averaging observation for the same year as described in point 4.3) of each one of the 24 Port brands analyzed. Panel B shows summary statistics for the distributions of prices for each brand in US Dollars.
From the analysis of the table we can easily understand that there exists a big difference in number of observations between the several brands, having the most observed brand which is Taylor’s 346 observations, while the least observed brands Borges, Dalva and Krohn have 2 observations each. This might happen because of reputation of each brand or rarity of the wine and in our case this clearly divides our sample according to quantity of sales observations as almost 50% of our observed brands have over 100 observations. The average of observations for all brands is 113,63. If we take into account the yearly observations for the all sample we conclude that the year with more sales observed is 2006 with 198 while the year with fewer sales is 2000 with 82. The average of number of annual observations is 144 for the 19 years which are part of our sample.
In line with the difference in number of observations the hammer prices for each brand also show a very big discrepancy between brands. This also indicates that some brands are more appealing and therefore have a greater ease of being transacted through specialized auctions and achieve higher hammer prices. Analyzing the means for our observed brands we also conclude that this measure clearly divides our sample in 2 with around 50% of our sample having a mean superior to 100 USD. In our sample the highest hammer price achieved is 9108,00 USD for Quinta do Noval Nacional while the lowest price observed is 4,01 USD for Croft, if we analyze the average prices for each brand we conclude that means vary from 27,21 USD for Kopke to 860,30 USD for Quinta do Noval Nacional, while the general mean for all sales included in our analysis is 240,50USD, however this value should be used with caution for any possible detailed analysis due to the differences in number of observations for each brand.
6.2.4. Vintage Port index vs Portuguese stock index 20
In line with previous studies as Lucey and Devine (2015) or Dimson, Rousseau and Spaenjers (2015) between others we decided to compare our calculated vintage index Port with an index that represents the capital market, in our case we decided to use the Portuguese stock index 20 or PSI20 as this represents the Portuguese market, in this specific case the index is based in the performance of the top 20 companies in Portuguese stock market, or in different words it is constructed by shares of the 20 highest ranked companies listed on Euronext Lisbon in terms of free float market capitalization. Data to compute PSI 20 index for comparison was obtained from Pordata which uses as data source “BP – Securities Issues Statistics†and adjusted to reflect the same period as our analysis.
6.2.7. Sharpe ratio for calculated indexes
In order to measure how our calculated indexes perform relative to their levels of risk, we decided to follow the same approach as Lucey and Devine (2015) and compute the Sharpe ratio for each of the indexes. We use our calculated data for average returns of each index and standard deviations, for the risk free rate we follow the same line as when we choose PSI20 and we use the average rate for the Portuguese treasury bonds of 10y for the same period , even that these show higher risk than others.
Sharpe ratio is calculated as follows (Lucey, 2015):
Sharpe = r ̅_(p- r_f )/σ_p (4)
Where r ̅_(p ) represents the expected Portfolio return, r_f represents the risk free rate of return and σ_p is the standard deviation of the Portfolio.
Table 8 shows the results for our calculated Sharpe ratios for each index
Index Average index returns Standard deviation Risk free rate Sharpe ratio
PSI20 0,0324240 0,2296630 0,0496667 -0,07508
Vintage Port 0,0442880 0,1042190 0,0496667 -0,05161
Y1963 0,06309 0,198751 0,0496667 0,067538
Y1970 0,070951 0,154655 0,0496667 0,137625
Y1977 0,049124 0,197527 0,0496667 -0,00275
Y1985 0,036776 0,123599 0,0496667 -0,10429
Y1994 0,033939 0,253461 0,0496667 -0,06205
Cockburn's 0,062286 0,174034 0,0496667 0,072511
Croft 0,160925 0,552824 0,0496667 0,201255
Dow's 0,091752 0,142264 0,0496667 0,295826
Fonseca 0,134971 0,390852 0,0496667 0,218252
Graham's 0,008215 0,14733 0,0496667 -0,28135
Noval 0,114955 0,351873 0,0496667 0,185545
Noval Nacional 0,089003 0,190746 0,0496667 0,206224
Sandeman 0,084154 0,379632 0,0496667 0,090844
Taylor's 0,026902 0,132038 0,0496667 -0,17241
Warre's 0,057105 0,18507 0,0496667 0,040192
Table 8: Sharpe ratio for each index
Analyzing our results for the calculated Sharpe ratio, 7 out of our 17 calculated indexes show a negative Sharpe ratio meaning that those indexes performed worse than the risk-free asset used for calculation. In these cases the return for investor would be better for the period if the bet was in the risk free asset. In this group it is included our vintage Port index, Graham’s and Taylor’s indexes and the annual indexes for 1977, 1985 and 1994. The lowest Sharpe ratio is for Graham’s of -0,28. Regarding our vintage Port index even returning a negative Sharpe ratio it is still higher than the PSI20 negative ratio.
Regarding the remaining indexes all show positive ratios, however always lower than 1. In this group the lead is for Dow’s with a ratio of 0,2958 followed by Fonseca, Noval Nacional and Croft all with ratios superior to 0,2. All other calculated indexes have results between 0,04 and 0,19.
These values can be seen as a result of the high volatility that makes these assets risky with similar average returns to the risk free asset.
7. Conclusions
Through our analysis we are able to conclude that our vintage Port index and most of our brand and year calculated indexes show a better overall performance than the index PSI20 chosen to serve as comparison to the market over the analyzed period, this goes in line with previous studies as Masset and Weisskopf (2010) and Lucey and Devine (2015) as our indexes show generally higher returns and lower volatility than the market. However we also need to notice that the volatility for these indexes is generally high as can be easily concluded by the results obtained in the Sharpe ratio calculation, meaning that the investment carries high risk without enough compensating returns in some of the cases. The high volatility and the returns in no excess of the risk free asset are a setback to possible investment decisions, however there are some exceptions, suggesting that an excess return might be a possibility in some particular cases.
Adding to these market risks we also need to highlight the risk of breaking bottles or wine to spoil as it is a perishable asset and normally a low percentage of bottles is even expected to perish during its life, returning a total loss to investor.
Our results also suggest that brand might be a more determinant factor for price definition than the vintage year as our calculated top brand indexes seem to have better results than the vintage year indexes.
From the small number of observations gathered compared to other studies we also conclude that the great illiquidity of wine market is also a drawback for this kind of investments in comparison to others and can serve as a barrier for these investments to be considered interesting. The simplicity and fastness of the trades nowadays in the capital market turns the wine market illiquidity obvious. This great illiquidity also counts as a hidden cost according to Dimson and Spaenjers (2014).
In terms of costs and as most of the sales are done through auctions we also conclude that transactions costs tend to be high for wine as for all emotional assets trades, fees and mark ups from auction houses and dealers on collectibles are high, easily achieving more than 25% of the assets price as it is stated by Dimson, Rousseau and Spaenjers (2015).
Adding to transaction costs, holding costs for investing in wine can become high as the collector or investor has to keep the quality of the asset and in the case of wine as it is a perishable item this means that the bottles need to be storage in very specific conditions or pay a yearly fee for the wine to be storage professionally.
All these costs can also be a drawback to a possible investment decision, however we decided not to include any cost in our analysis as there is no common line across the hypothetical costs for transactions in literature and also we did not account transaction costs for the market.
With all these drawbacks and following our results we conclude that wine is a risky and very illiquid asset, however we highlight that there is the possibility of specific Portfolios to deliver interesting returns under certain specific conditions.
During our research and even through our analysis we are in constant contact with other possible points of interest which might be later used as future research topics mostly related to our main theme, between those we can think of analyzing returns for other types of Portuguese wines such as Madeira wine, other type of Port wines as tawny or Port rarities and also for other Portuguese wines. We can also think into all the aspects that drive the prices of these wines to move as for example the renowned critics power on price definition or the impact of the yearly weather on prices, we can even try to predict the future returns of a specific wine according to several possible inputs in a predefined model.
To finalize and due to the enormous prestige and quality of Port wine worldwide we must always leave open the option that the enjoyment of the consumption in some cases turns that value to exceed the value of storing this wine for possible returns.