Featured Predicting hammer price: playing a bit with Sixbid data & looking for inspiration

I would not lump all auction houses together.

Take for example an auction house known for taking photos that make the coin look better than it really is, and selling coins with undisclosed problems. Then as another example an auction house known for premium quality coins that are frequently better than depicted, with conservative grades given. Both examples I mentioned may sell the same coin with the same listed grade, however, the better auction house might have the coin hammer for 10x as much. When taken together along with average auction houses, these can look like outliers, but they may not be outliers at all for those respective auction houses. Your data for an average house may be a much tighter distribution, and for the outliers, you may have auction houses that overgrade, and others that undergrade.

It may be useful to examine each auction house to determine how to handle them. For example, the better auction house in my example sells better coins. You may be more inclined to bid at that auction house for this reason, however, your data taking into account all auction houses will result in you underbidding at this particular auction house, and that isn't a helpful extrapolation from the data.

 

Log in or Sign up to hide this ad.

@Ken Dorney I'm not quite sure I understand your histogram. What does the Y-axis represent? A cumulative revenue? There does seem to be a trend over time, but to sort of get a feeling whether there is a likely true increase/decrease compared to the previous month, statistical testing comes in place. If I understand it correctly, the Y-axis is a mean (?) of a couple of (?) years (?). If so, a simple T-test would suffice to get a general feeling. If you're comfortable enough in providing the data, I can do this.

@rrdenarius: you're probably completely right: the number of bidders is likely to say something about the hammer price. I do not have this variable however. In my experience, it is only visible during the auction, and not after the auction. If looking at causality (I've tried four case studies so far: the two being shown here (JC denarius; Augustus divus Caesar denarius), a medieval Anglo-Saxon penny, and the Attica Tetradrachm), there certainly is a very strong significant relation between grade and estimate, even when corrected for other variables such as toning y/n, damaged y/n, provenance, etc. This is to be expected of course, as I've explained above.

@Jaelus it would probably be a fun exercise to compare estimate vs hammered of each auction house and add this as a weight to the model. I'll consider it, but it's of course a lot of work and the estimate will probably not be very precise. Still fun to consider.

For those still following the thread, I'm trying a couple of things:
1) predicting the hammer price (i.e. the price a coin will sell for) given various variables
2) establishing a relation between these variables and the hammer price

Please note that in model 1 (prediction model) a non-causal relation is implied, while in model 2 (aetiological model) an causal relation is implied.

For this, I am looking for information that might be related to the hammer price. So far, I've got this:
- estimated price
- grade
- provenance
- NGC or other
- broken / damaged / off center / repaired / scratched, etc.
- bankers mark
- month of the year
- (auction house might be a possibility, but is difficult due to the number of auction houses per given coin and thus the required flexibility of the data for that number of degrees of freedom).

Do you have other possible predictors?

 

The other reason to track and weight by auction house is that some auction houses are terrible at estimating, at least for some material. For the areas I collect, I can point you to auction houses that estimate with a high degree of accuracy, and others that will estimate 50 Euro on a coin that hammers for 900 Euro.