Addressing Near Perfect Correlations In Highly Competitive Commodity Markets

I am working on an analysis for a highly competitive agriculture market. I'm looking at county elevators in Iowa where competitors set their price for tomorrow's corn "ask" the evening before, but after the market closes. The markets close around 4PM here and the elevators set their prices around 7PM. There are 5 competitors in this market. As an example, if the price increases between t & t-l (where t is today's closing price) is \$0.50 per bushel at a regional market, each elevator will raise their asking price for bushels by some percentage of the increase. That amount has an upper limit to prevent arbitrage, but for simplicity's sake, assume they can capture the full \$0.50, but often don't choose to because they want to sell the most corn possible and not price themselves out of the market. My goal is to understand how competitors have behaved historically. Have they tried to capture the full price increase or did they capture a smaller percentage of the increase?

I took the analysis over from a colleague who tried regular regression to model the data. As a starting point, I'm using two-stage least squares since there are some real endogeneity issues. However, are there any better tools which could be useful? The once which comes to mind is a random forest because of the relative importance tools, but is there a way to extract actual percentages? I'm careful to not use the word coefficients. Thanks in advance.

• Could you describe the data? What do they consist of, exactly? – whuber Mar 26 at 21:40
• They are all differentials to the main market hub prices. – Confucion Mar 26 at 21:41
• A lot of information missing. On the statistics side, what was the regression run, what you believe the endogeneity issue is, what IV was used. On the economics side, is this a two sided auction, how does the demand side trade (only supply side is described in the question), do these "elevators" make markets on both sides, does the demand side also submit bids, if so do you have the bid-ask spread data and how does it look, etc. – Michael Mar 27 at 0:01
• Michael, agreed. No IV has been tried yet. I'm still trying to figure out which functional form is possible to get unbiased results. 2SLS seems like the most obvious from a regression standpoint. Elevators make a market on both sides, but I'm only focusing on the supply side at this point. I do have the B-A data and it's roughly what you'd expect - typically for corn its \$0.25-0.50/bushel. – Confucion Mar 27 at 13:26