Postal delivery pricing: econometric analysis
In November 2018, the Scottish Government launched the Fairer Deliveries For All: An Action Plan which listed eight key actions to tackle the unfair and discriminatory parcel delivery charges faced by communities in remote and rural Scotland. This report is in response to action points 1 and 2.
This section describes the approach of our econometric analysis, covering data collection, model specification and estimation.
Our research covers a large share of postal delivery and online retail markets in Scotland and explores all postcode sectors in Scotland. Consequently, our overall results provide a good indicative picture of what drives delivery prices and how they vary across different postcode sectors in Scotland. However, it should be noted that our sample is not a true random sample of all deliveries to Scottish addresses and thus it does not provide a fully representative picture of delivery charges in Scotland. Similarly, while our model captures key factors affecting prices, including parcel characteristics, delivery speed, type of service, areas’ rurality and remoteness and what company delivers each parcel, there are some interesting area and company characteristics that due to data restrictions were excluded, e.g. fuel costs, labour costs, and companies’ business models.
Our data collection process followed a four-step approach:
1. Postcode selection: We created a list of all 1,029 postcode sectors in Scotland, randomly selecting a full postcode for each of the postcode sectors. Postcodes that have been discontinued (per the Scottish Postcode Directory) were excluded from the selection process. We focused on the sector level in our analysis as preliminary analysis of data provided by Resolver indicated no variation across full postcodes within a single postcode sector.
2. Postal operator selection: The selection of delivery companies included in our sample was informed by the market research exercise we undertook during phase 1 of the project to identify the biggest delivery operators across the UK (see Box 6, The UK delivery market).[22, 23]We focused on large national operators with widespread name recognition, as most individuals will rely on these companies to ship products on a regular basis and most retailers will contract these companies for product delivery. In addition, large firms were more likely to have a price quote service on their website as opposed to calling the operator or requesting a personalised quote via e-mail, which ensures prices do not depend on characteristics other than delivery location, speed or size.
We included six delivery companies in our dataset: Parcel Force, My Hermes, DPD, TNT, Yodel and Menzies. Menzies, in contrast to the other five operators, is a regional company that only delivers to postcodes in the Highlands and Islands and Aberdeenshire. These operators differ in the delivery options they offer to customers, outlined in Table 1.[25, 26]
Box 7: The UK delivery market
We undertook a market research exercise to identify the key couriers and delivery companies operating in the UK market. In terms of market share by parcel numbers, Royal Mail accounts for 47% of the UK market, followed by Hermes with 11% and Yodel with 8%. The remaining 34% is shared among the other UK couriers.
When considering total revenues, the landscape is slightly different; Royal Mail remains the market leader followed by DPD, UPS, DHL and TNT.
Top five biggest delivery companies in the UK, 2017 (share of total annual revenue)
Royal Mail – 35.0%
DPD – 12.0%
UPS – eight per cent
DHL – seven percent
TNT – six per cent
Alongside the delivery companies that operate UK-wide, there are some additional players important for the Scottish market such as Menzies and CalMac. Menzies is a delivery company used by national operators for last-mile deliveries in remote and rural areas in Scotland such as the Highlands and Islands. CalMac, on the other hand, is the major operator of ferry services between the mainland of Scotland and 22 islands on Scotland’s west coast.
For each postal operator, we defined three different package sizes to request quotes for:
a. Small parcel: length 45 cm, width 35 cm, depth 16 cm, weight 2 kg
b. Medium parcel: length 61 cm, width 46 cm, depth 46 cm, weight 20 kg
c. Large parcel: length 150 cm, width 50 cm, depth 50 cm, weight: 30 kg
|Parcel Force||2 days||2 days||Yes||Yes|
|My Hermes||2 working days||No||No||No|
|DPD||1-2 working days||No||No||Yes|
|Yodel||2 working days||2 working days||No||Yes|
|Menzies||No||3-5 working days||No||No|
3. Retailer selection: Similar to delivery companies, we conducted market research to identify key retailers in the UK market (see Box 7 above, The UK retail market). [27, 28]
Box 8: The UK retail market
To ensure that the sample of companies we used for our analysis was a good representation of leading retailers in the UK we also undertook market research to identify the key players in the sector.
Based on data from GlobalData, the “Big Four” supermarkets – Tesco, Sainsbury’s, Asda and Morrisons – accounted for 30.3% of the total annual revenue generated by the sector in 2017. Amazon was in fifth position and Marks & Spencer and John Lewis PLC jointly occupied the sixth position. In the UK, online sales account for around 15% to 20% of total retail sales – with this figure predicted to rise as technology is becoming increasingly embedded in the lives of consumers. In the online retail market Amazon, Tesco and e-Bay are the top three players comprising around 33% of total retail sales. Other big retailers include Asos, Argos, John Lewis PLC and Next.
We focused on large retailers that deliver across the UK, as most individuals are likely to purchase products from stores and websites they recognise. However, to ensure that our sample has sufficient variation – due to the existence of multiple players in the UK retail sector – we also included a number of smaller retailers. We selected one of three products – a two-seater sofa, a 43-inch television or saucepan set – as these are large items that are the most challenging to deliver to remote areas, and thus have the greatest variation in ability to deliver, delivery times and shipping fees. For each retailer and postcode sector, we collected the price for regular shipping and home delivery (as opposed to click-and-collect).
Based on our market research our sample was comprised of three of the UK’s key players; Argos, Asda and Sainsbury’s. However, given the plethora of retail businesses operating in the UK – more than 390,000 – we also chose a number of smaller retailers such as ao.com, Furniture Village, Hughes, Ikea and Sonic Direct, to ensure sufficient variation. Table 2 lists the retailers and the corresponding products included in our sample.
|Furniture Village||Two-seater sofa|
We also collected data for retailers that were ultimately excluded from the analysis due to incomplete information available online. For example, some retailers displayed the message, “Deliveries to your postcode may incur a surcharge; as such, online checkout is not available. Please call for pricing and delivery times”, which does not provide information on the magnitude of the surcharge or possibility of delivery.
One important challenge with collecting price data from online retailers is that they might cross-subsidise shipping expenses by incorporating part of the shipping cost in the product price. This means that the shipping fee listed does not necessarily represent the cost of delivering the package. Since it is difficult to precisely identify to what extent retailers follow this practice, we collected data for both delivery companies and retailers and treated pricing data from retailers with great caution.
4. Mystery shopping: To collect pricing data from delivery companies and retailers, we automated a “mystery shopper” approach using Selenium, an open-source web automation tool. We wrote a custom Python program for each retailer or postal operator to request shipping quotes from websites and combine them into a database for further analysis. Most retailers require a delivery address before providing the user with a shipping price estimate, while delivery companies also require entering package weight and dimensions. Our program automated the process of entering parcel characteristics and addresses and extracted each price quoted. For postal operators, we used as a sender’s address the Alma Economics’ London office address and for parcel weight and dimensions, we used the categories specified in step 2 (Postal operator selection).
Additional details on sample sizes for individual retailers and delivery companies can be found in Annex A.
Apart from prices, we collected data on postcode and carrier/retailer characteristics that were potential factors in per-parcel pricing models. Surveys of postal carriers to the Highlands and Islands have pointed to fuel costs and driver hours as one of the important contributors to higher costs. We cannot include these directly in our model due to the lack of available data, but we can include urban-rural classification and population density as proxies, respectively. Our full list of variables included as controls in our econometric models include:
1. Population density: This serves as a proxy for delivery volumes, as due to economies of scale we expect areas with more frequent deliveries to face lower postal charges. We collected this data at the postcode sector level from the 2011 Scotland Census, although no data was available for 87 postcodes for the following reasons:
a. The postcodes selected were introduced after 2011;
b. The postcode had a population of less than 20 households and/or 50 people; and
c. The postcode is only classified as part of Large User Postcodes.
To address this, we developed and tested two specifications for our econometric model, one without the 87 missing postcodes and one that does not include population density.
2. Islands: To account for increased costs of shipping due to ferry or plane requirements, we include island dummy variables (using the island identifiers found in the Scotland Postal Directory).
3. Urban-rural classification: We obtained this data through the Scottish Government Urban Rural Classification. There are several different types of classifications based on level of detail, and our main classification used is eight-fold: postcode sectors are classified first as a settlement or geographic area, then segmented by population and drive time to settlement (for areas). A full definition of the different categories is presented in Annex C.
The process of collecting both postal charges and other drivers of shipping costs produces a dataset where the unit of observation is a delivery request for a specific parcel from a postcode in Scotland to a specific company or retailer.
The aim of this work is to understand the main delivery cost drivers and the different factors contributing to the surcharges imposed in rural and remote areas of Scotland. As there is no available information on costs for either postal operators or retailers, we used delivery prices as a proxy. Each company has a profit margin which is added to the delivery cost to create the delivery price given to the consumer. Consequently, the price is not expected to directly reflect the costs, but it is a credible indicator.
While postal operators’ price is expected to mainly include the cost and a profit margin, for retailers the delivery price structure can be more complicated. More particularly, retailers often cross-subsidise delivery charges by incorporating part of the delivery costs in the product price. For example, many retailers offer free shipping to all postcodes for which delivery is available, although they do incur shipping costs.
Furthermore, all the retailers in our dataset offer the same delivery price everywhere. That said, some retailers do not deliver to specific postcodes at all, which offers an insight into whether retailers discriminate or not against specific areas in Scotland.
Taking the above considered into account, we decided to undertake two separate pieces of analysis:
1. Factors affecting postal operators’ delivery prices
2. Factors affecting retailers’ probability to offer delivery
Model specification and estimation
It is difficult to directly compare postal charges across different postcodes, as operators vary their offerings along several dimensions and many different factors drive differences in charges. To isolate price changes from differences in fuel costs, driving times and delivery volume, allowing for postcode sector comparisons on a like-for-like basis, we use an estimation method called hedonic pricing analysis. This approach is based on the premise that the value of a good or service (in this case, the price of shipping a good to specific Scotland addresses) is based on its characteristics. Using different regressions, our econometric model aims to estimate the average delivery price for each postcode/address based on area and parcel characteristics.
Our primary specification for delivery companies is outlined in Equation 1:
with representing each postcode sector, j each parcel size and h each postal operator/retailer. In this equation:
- is the postal charge in pounds quoted by a postal operator to deliver a package of size to a randomly chosen postcode in postcode sector .
- are area characteristics, including dummies for local authorities and postcode sectors, postcode sector population density, urban-rural classification and island dummies. These serve as a proxy for delivery volume or density; we expect economies of scale in more densely populated, urban postcodes as multiple parcel deliveries can be arranged through one journey, lowering the price of delivery for each parcel.
- are parcel characteristics, including parcel size, delivery time and location (drop-off or door-to-door).
- is a vector of dummy variables for each postal operator. This allows us to identify whether certain delivery companies charge higher postal charges on average compared to other companies, which may indicate higher mark-ups or greater inefficiencies.
- is an error term that reflects unobserved pricing drivers for each observed price
The above model was estimated using ordinary least squares. Each coefficient () is an estimate of the impact of the area, parcel and company characteristics on postal charges. Using the dataset collected, we estimate several variants of this equation. Since there are very few postcodes for which operators refuse to deliver, we focus on price as our dependent variable.
Similarly, for retailers only, we estimate Equation 2:
As explained above, there is no variation in prices across different postcode sectors for a specific retailer – retailers in our dataset charge the same shipping fee irrespective of delivery location but often exclude certain postcodes from online home delivery. Thus, we include a binary variable for availability of home delivery as our dependent variable: 1 if the retailer delivers to a postcode and 0 if home delivery is not available.
The resulting estimation is a linear probability model, which allows us to estimate the probability that retailers offer delivery to a specific postcode sector (as opposed to the average postal charge). These coefficients are straightforward to interpret – in this case, each captures the marginal impact that each characteristic has on the probability of delivery availability for a certain postcode.
There is a problem
Thanks for your feedback