REGIONAL ANALYSIS

An analysis of the effect listing location has on expected price. De-jargon tooltips are marked in red. All links to sources or more information are underlined.

Introduction

The price of a commodity being sold on a listing on any given day can vary depending on where it is located. For example, Scotland contains a large number of mills and livestock farms, which means Feed Wheat is normally priced higher than we see in England. For example, below find the price of Feed Wheat for listings in the West Midlands compared to prices in North East Scotland.

We can see that North East Scotland achieves a consistently higher price over time than the West Midlands.

This analysis will investigate the effect regional premiums has on the price of bids. Including the following questions:

How does regional prices compare to the LIFFE price?

If we wanted to display a single price to users that represents the likely price they will get when listing a commodity in a certain region of the UK, how would we do that?

If we can find a correlated value for price which constantly tracks with the regional premium over the course of the year, we could adjust that price to get an estimate of the regional price on any given day.

We can get a consistent source of pricing data from the LIFFE/ICE Exchange. And we know this is strongly correlated with the price offered on Hectare Trading. So can we this relationship to calculate regional carry?

Let's compare the price of feed wheat offered on Hectare Trading, to the LIFFE price for feed wheat on the same day for North East Scotland and the South West.

Region Gradient Intercept

Looking at the equation of these regressions, they both have the same gradient of 0.96, which is similar to what we have observed previously. The however the intercept of the line is different for the two regions, with the South West having an intercept of 14.9, and the South West 5.16.

This shows that the regional premium does not scale with price (i.e. higher prices do not achieve greater regional premiums) and can be thought of as a flat adjustment (aka a weighting) that can be applied regardless of the price.

What this means in means in non-statistics speak, is that if we wanted to predict the price of Feed Wheat in North East Scotland based on it's correlation with the LIFFE price, we should take 96% of the LIFFE price and add a regional premium £14.90 to it. In comparison to the South West, where we would add £5.16.

Let's apply the same calculation to each region, see below the predicted price for each region, based on if the LIFFE was displaying a price of £200 a ton.

From these predicted prices we can see a pretty consistent divide between Northern England & Scotland and the Southern parts of England. With Scottish prices being particularly high compared to prices in England.

However, while these prices correlate, this prediction is pretty simple. Because we look at all price points, the correlation is a sort of average of the distribution. This works well when the relationship has low variation, i.e. the price difference between ICE/LIFFE is always the same.

Summary

Let's look at the amount of variance in the difference in prices per region to see if these predictions will be accurate.

How large is the variance from LIFFE/HT regional price?

Let's plot the distribution of each bid's price difference to the LIFFE price offered when the bid was made split by region. Prices that were lower than the LIFFE price that was available at the time of the bid will fall to the left of the chart, and those higher than the LIFFE price will fall to the right.

We can see the trend we observed still holds too. That on aggregate Scottish and Northern regions achieve a better price on Hectare Trading compared to the LIFFE. However we can also see that there is a very large amount of variance in the distribution.

For the majority of regions, the standard deviation is over 80% of the mean. In stats land, we would deep this as an extremely high variance. That means that there is probably another variables at play that is effecting the difference in price from the LIFFE from bid to bid.

This could possibly be due to:

Summary

How does regional premiums vary with seasonality?

Since the West Midlands has the most amount of data points, we will use as our group to investigate the seasonal effect on the regional premium. As we know Scotland has a distinctly different relationship vs the rest of the UK, We'll also monitor North East Scotland.

We can see a similar pattern here to our regional analysis. Prices in Scottish regions are higher on average than Midlands prices throughout the year. However, both sets have a high level of variance that makes it difficult to predict the exact price difference from the LIFFE on any given day.

Summary

How do regional premiums vary between buyers?

Another source of this variance could be down to how different buyers interact with certain regions. Especially as a large proportion of our bids come from buyers based in Scotland.

Could these buyers be skewing our results by offering significantly lower prices compared to the LIFFE, due to the fact they have to haul the purchased commodity a further distance?

First let's get the overall picture, by plotting the bid/LIFFE price difference split by the buyer who placed the bid. We'll do this for 3 regions, central and north east Scotland, as well as the West Midlands.

There are some key takeaways we can draw from this:

Firstly, we can see that even our English based buyers, such as Robin Appel, bid with a wide variation in price in the West Midlands. In the same regional weighting range (~ -£1) as we see when comparing across the whole region.

Secondly, we can see that when Scottish and nation-wide buyers are bidding in England, they are following a similar pattern as English buyers. Bidding generally £1-£5 lower than the LIFFE price in the West Midlands, with a high amount of variance bid-to-bid.

Summary

Can we create larger regions for a broader weighting?

What if we divided England into three districts, South, Midlands and North, as well as Scotland, and check to see if there is any clearer weightings with broader groups.

Let's divide Great Britain like so:

And then look at the price difference by GB Area.

We can see that while there are obviously other factors at play in pricing, listing in the South are normally priced lower than the LIFFE price (on median £5 lower), with Scotland being weighed extremely higher than than rest of Great Britain.

How does haulage effect bid prices?

The other factor that could be causing this variation is due to the cost incurred on the buyer in transporting the grain from farm to the buyer's storage location. As a buyer, the further away the farm from your storage location, the more you'll have to pay to transport it.

This is known as the haulage cost of a given listing. Normally this cost is hedged against the price of the listing. I.e. If the buyer is further away it will take into account this price by offering a lower price.

The high variance in bid price per region could be due the different haulage costs each buyer sets on a listing-by-listing basis. If we could quantify this haulage cost, we possibly could better normalise regional price predictions more accurately.

We'll attempt to do this by looking at three factors:

Distance between listing and buyer organisation

Let's first look at our most complete dataset, the distance between where the listing was created, and the location of the buyer's designated organisation. This would be the easiest option to create a haulage factor with, as each organisation has to provide us a primary location.

For example, below we can see the distance between GrainCo's primary location and the location of the listings they bid for.

Now let's take the difference from the LIFFE price for each bid, and compare it to the distance. For example, the highlighted bid between Grainco and was km away, was £ above the liffe price.

If we plot this for all of Crainco's bids, we can see the relationship we expect, the further away the bid is, the lower the bid price is in comparison to the LIFFE.

The relationship between the organisation's location and the listing location, and the price offered has a weak negative correlation, which has a statistically significant p value.

Other large buyers also follow this trend. Cefetra for example, has a negative correlation between price and distance to listings.

However, not all large buyers follow this trend. Robin Appel for example has a very weak, verging on non-significant correlation in the other direction. Meaning they would be paying more for crop that was further away. 😱

This is most probably due to the fact that for the majority of our buyers, the location of the organisation's offices is not where the grain is actually being moved to. Which is what is deciding haulage costs.

If this is the case, while we'll get a general idea of the haulage, the difference between the organisations HQ and the actual destination for transported crops could be considerably different.

Distance between the listing and the nearest storage location

So can we estimate where the listing's commodity is being hauled to and see how this effects price? For example, Grainco list where they store and dry their grain here. Let's estimate the location of these storage centres, and then calculate the distance from each listing to the nearest Grainco store.

Again, let's plot each price's difference from the LIFFE at the time the bid was placed, this time against the difference to the nearest Grainco Store.

While there is a similar relationship, it has not improved our correlation or reduced the variance from listing to listing. Let's just check if these true by checking another one of our large buyers, Cefetra.

From a quick internet trawl, we can find a number of their major storage sites across the country. Shown below:

Again, lets compare each listing to the nearest major Cefetra grain store.

Again, this doesn't improve our correlation or reduce the variance. This is probably due to the fact that we will be missing some Cefetra's grain storage locations in this very basic list gathered from online sources.

In conclusion, without adequate/complete storage information from buyers, it will be very difficult to establish a firm prediction on the effects of haulage on price.

Distance between the listing and the nearest ICE storage location

Grain that is going to be sold on the ICE exchange, needs to be stored in an official ICE exchange storage location. A list of these can be found here.

Is there a relationship between the price offered by the buyer and the distance the listing is from one of the ICE locations? As the ICE storage locations are only listed in England, we'll only look at bids placed for listings based in England for this comparison. Press the button below to select a random listing and see it's nearest ICE storage location.

Now let's plot this for each bid made in England.

Unfortunately, there isn't any significant correlation between the distance of the listing and the nearest ICE storage location. Again, this could be due to incomplete ICE storage data. However, currently we would not be able to estimate the effect of haulage on prices based on the distance of the listing to the nearest ICE storage location.

Summary

Conclusion

To summarise the results of the analysis:

Regional Premiums

Haulage

Recommendations