recentpopularlog in

jerryking : sotheby’s   1

Art market ripe for disruption by algorithms
MAY 26, 2017 | Financial Times | by John Dizard.

Art consultants and dealers are convinced that theirs is a high-touch, rather than a high-tech business, and they have arcane skills that are difficult, if not impossible, to replicate..... better-informed collectors [are musing about] how to compress those transaction costs and get that price discovery done more efficiently.....The art world already has transaction databases and competing price indices. The databases tend to be incomplete, since a high proportion of fine art objects are sold privately rather than at public auctions. The price indices also have their issues, given the (arguably) unique nature of the objects being traded. Sotheby’s Mei Moses index attempts to get around that by compiling repeat-sales data, which, given the slow turnover of particular works of art, is challenging.....Other indices, or value estimations, are based on hedonic regression, which is less amusing than it sounds. It is a form of linear regression used, in this case, to determine the weight of different components in the pricing of a work of art, such as the artist’s name, the work’s size, the year of creation and so on. Those weights in turn are used to create time-series data to describe “the art market”. It is better than nothing, but not quite enough to replace the auctioneers and dealers.....the algos are already on the hunt....people are watching the auctions and art fairs and doing empirics....gathering data at a very micro level, looking for patterns, just to gather information on the process.....the art world and its auction markets are increasingly intriguing to applied mathematicians and computer scientists. Recognising, let alone analysing, a work of art is a conceptually and computationally challenging problem. But computing power is very cheap now, which makes it easier to try new methods.....Computer scientists have been scanning, or “crawling”, published art catalogues and art reviews to create semantic data for art works based on natural-language descriptions. As one 2015 Polish paper says, “well-structured data may pave the way towards usage of methods from graph theory, topic labelling, or even employment of machine learning”.

Machine-learning techniques, such as software programs for deep recurrent neural networks, have already been used to analyse and predict other auction processes.
algorithms  disruption  art  art_finance  auctions  collectors  linear_regression  data_scientists  machine_learning  Sotheby’s  high-touch  pricing  quantitative  analytics  arcane_knowledge  art_market 
june 2017 by jerryking

Copy this bookmark:





to read