supplementary material file

Presentation of the Xls. File “supplementary materials” This file is organized by sheet as follow: Sheet Rdatatime : Contains the data used for the R code. All hedonic and FMM calculations can be replicated using the R code applied to the Rdatatime matrix. Sheet Coding: Contains the names of the variables associated with each X-number code. Sheet Sales number: displays the number of auction lots (each lot containing one watch), broken down by quarter on the left side and by half-year on the right side, for the entire sample and by location (New York, Geneva, Hong Kong). Sheet Appendix 1 As prices include the buyer's premium—whose rate varies according to the hammer price level—we must account for its potential effect on watch prices. This sheet presents the data used to build the figure in Appendix 1. On one hand, for each auction (place and date in column A), the minimum (E), maximum (F), first quartile (G), median (H), third quartile (I), last five percentiles (J), and mean (K) prices are calculated. On the other hand, we identify the two fee thresholds above which commission rates change. Note that each threshold is initially expressed in local currency. Columns C and D (in green) contain Fee Thresholds 1 and 2, respectively, converted into USD using the exchange rate on the last day of the sale. Columns M, N, and O (in grey) contain the commission rates associated with each fee threshold. Sheet Rarity construct details the process used to build the rarity variable. For each observation (identified by the same ID as in Rdatatime, column A), we extract the expert comment (column B). First, we extract from the comments the terms related to watch rarity (rarity, rare, scarce, scarcity, unusual) using the following function (column D): =SI(OU(ESTNUM(CHERCHE("rarity"; B2)); ESTNUM(CHERCHE("rare"; B2)); ESTNUM(CHERCHE("scarce"; B2)); ESTNUM(CHERCHE("scarcity"; B2)); ESTNUM(CHERCHE("unusual"; B2))); 1; 0) Second, we manually identify the intensifiers of rarity terms, as these are not always adjacent to the target term in the comment or separated by a comma. The intensifiers are then detected using a SI(TROUVE) function. After manual verification, the coding of intensifiers (column F) is incorporated into column D, and their descriptive statistics are presented in columns G to K. From a semantic perspective, we assume that when an expression reads “intensifier term1 and term2”, the intensifier is understood to apply to both terms. Sheet Liquidity traded details, step by step, the calculation of Equation (4). The final output is shown in column N, and its Box-Cox transformation is presented in column O. The watch brand is identified in column G, and its reference in column C. Note that the cumulative observation numbers (Cum) for reference and brand have been shifted by one unit to avoid a zero denominator. The lagged values of Cum ref and Cum brand (t−1) have been computed using a dynamic cross-tabulation and assigned via Power Query. Sheet Liquidity price details, step by step, the calculation of Equation (5). From column B to column F are the calculation steps for price-based liquidity, In columns H to J (in blue) are the necessary transformations for applying the Box-Cox transformation to the variable. Sheet Reputation construct The monthly Google Trends score for each brand appears in blue for the term "watch" associated with [brand], and in green for the term ""montre"" associated with [brand]. The salmon-colored blocks show the weighted average, calculated as 60% for the "watch" score and 40% for the "montre" score, based on the market share of each location. On the right side of the table, you’ll find the average of the monthly weighted scores by quarter for each brand. We assigned these quarterly averages to the dataset using a dynamic cross-tabulation and a Power Query assignement for each brand–quarter pair. Sheet contextual variables presents the MSCI world and Bitcoin series in current US$ Sheet Figure 4 predictors’ contribution to adj-R² shows the contribution of each predictor category to the explanatory power of the model.This analysis is relevant for the replicability of the hedonic index using a reduced information matrix. Sheet WS Hedonic Estimate Quarter presents the robust estimation of the hedonic model, incorporating both stochastic and structural characteristics, for the whole sample (WS) using a quarterly time frequency. Sheet WS Hedonic Estimate Half-Year presents the robust estimation of the hedonic model, incorporating both stochastic and structural characteristics, for the whole sample (WS) using a half-yearly time frequency. Sheet Number of Clusters presents the results of the information criteria used to determine the optimal number of clusters in the Finite Mixture Model, along with the number of observations in each cluster.The information criteria are computed for different clusterings, ranging from 1 to 5 clusters. Sheet EM Outputs Flexmix 3s/4s exhibits the estimated coefficients for the models with 3 and 4 clusters. The arguments provided in the manuscript support the selection of the 3-cluster specification. The coefficients assigned to the profiling variables (in bold) allow us to identify the type associated with each cluster. Sheet Hedonic estimate collectors (S=1) Sheet Hedonic estimate Amateurs (S=2) Sheet Hedonic estimate investors (S=3): present the estimates of the hedonic coefficients based on data filtered by cluster identifier.Cluster identification is obtained from the flexmix output, according to the highest posterior probability that an observation belongs to a given cluster. With the exception of S = 1, for which the Breusch–Pagan test is not significant, all estimations are robust to heteroskedasticity. Sheet estimated coefficients contains the quarter estimated coefficient of each hedonic function. Note that, hereafter, only quarterly coefficients are estimated for the cluster-specific indices. Sheet Mean price indices (Q/S) calculates the mean price index used for comparison with the hedonic index, based on prices from the Rdatatime dataset. This mean price index is not a market-wide index; it applies only to the Phillips dataset. In addition to the mean, the sheet reports the median, first and third quartiles, maximum, and minimum prices. Mean price indices are computed at both quarterly (Q) frequency (on the left side of the sheet) and half-yearly (S) frequency (on the right side). The indices are visualized in graphs at the far right of each table. Each mean price index (Q and S) is smoothed using a third-degree polynomial. Sheet hedonic indices quarter: provides the hedonic indices for the Whole Sample (WS), as well as for the Collectors, Amateurs, and Investors clusters. Columns B to E contain the estimated quarter coefficients taken from the “Sheet Estimated Coefficients”. The implicit prices are computed in columns F to I, using the parameters shown in columns J and K, according to Equation (2) of the manuscript. Then, columns M to P present the index calculations based on Equation (4). Missing values due to infrequent sales are linearly interpolated. Each index is visualized in graphs to the right of the table. Based on each index, the corresponding return series is calculated and displayed on the far right of the sheet. Sheet hedonic index half-year contains only the Whole Sample hedonic index and the corresponding return series at a half-year frequency, for comparison with the quarterly Whole Sample hedonic index. Sheet Collectible and conventional index exhibits the various benchmark indices used for comparison with the watch indices.A note at the end of the table describes each index along with its origin. Note that each indicator has not the same reference period and Gold, MSCI SP500 and FTSE WGB are expressed in US$. Sheet financial indicators. All indices are presented in columns C to Q and their return series in columns R to AF. The VIX appears in the last column (AG). The sheet is divided into two parts. In the upper part, benchmark indices have their real values for each quarter. In the lower part, we replace the values of the benchmarks for each quarter where the watch index has missing values by an interpolated value. Indeed, the comparison of volatility between indices requires that the same approximation be applied to all benchmarks.We calculate the financial indices only on the lower part. All indices are in formula except the drawdown maximum according to its sequential process. The risk-free rate retained is a mean on the period 2015 2023 (US treasury bonds from bloomberg) Sheet Figure 9 benchmark comparison: provides the graph corresponding to Figure 9, which compares the mean return and return risk of the watch indices with benchmark indices. END