Valuing the Environmental Benefits of Canals and Canal Restoration Using House Prices

abstract:This article values the environmental benefits of historic, navigable canals using property values. We improve on standard methods by controlling for microgeographic fixed effects and applying a difference-in-differences method to canal restoration. We find a localized price premium within 100 m, around 5% before the 2008 recession, dropping to 3.4% by 2016. These effects are driven by urban canal-side properties with a direct outlook on the canals or immediate access. These locations are also attractive for developers, with a higher proportion of new-build sales. Our estimates suggest that canals generate land value uplift of £0.8–£0.9 billion in England. (JEL Q51, R21)

3 4 bodies in general (Nicholls & Crompton, 2017) much of this literature relates to rivers, streams and other natural water features. Some of this work looks at the impact of individual cases of river restoration and improvement (Streiner & Loomis, 1995), including dam removals (Lewis & Landry, 2017) and the planting of riparian buffers which obscured river views (Mooney & Eisgruber, 2001).
These studies suggest that river improvements, views of rivers and reduced risk of flooding are all valued by homeowners, although they shed little light on the value of the amenity value of manmade canals specifically.
The literature on canals is much more limited, consisting of a few cross-sectional analyses, often on small samples for specific cities. These suggest that properties close to canals attract a premium, 2.9% for canal-side properties and 1.9% for properties within 200m in London (Garrod & Willis, 1994), 0.074% per metre closer to canals in Milan (Bonetti et al. , 2016) and 11% for properties with a canal frontage in Texas (Nelson et al. , 2005). Although the literature consistently reports that households pay a premium to reside near canals and other types of waterway, the magnitude of the estimates varies across studies and contexts, calling in to question the external validity of studies carried out on specific cases (Lewis & Landry, 2017).
Our work offers several contributions to this and the wider environmental evaluation literature.
Firstly, the size of our administrative data set on the universe of transactions in England and Wales means we can restrict our estimation sample to properties with 1500m of canals and estimate from variation in distance within these buffers, while still retaining more than 2 million transactions.
These data cover a wide range of urban and rural contexts, improving the generalisability of the findings. Secondly, by merging price transactions data with a rich set of geographic and socioeconomic data sources we are much better able to control for, and test sensitivity to, land use, distance to geographical features, employment and demographic variables. In our preferred specifications, we further control for fixed effects at a small geographical scale either Middle by guest on November 13, 2022. Copyright 2021 5 Layer Super Output Areas (MSOAS) or Lower Layer Super Output Areas (LSOAs) and for differing price trends at Local Authority District level. This means we estimate the price effects from variation in the distance to canals, and associated variation in house prices, that occurs within these small geographical areas. Confounding factors that vary at a higher geographical level such as access to labour markets are eliminated.
Lastly, we exploit the natural experiment of the restoration of a canal in England in a difference-in-differences analysis. The difference-in-differences method is frequently applied in valuing amenities/disamenities such as proximity to transportation nodes (Gibbons & Machin, 2005), wind farms (Gibbons et al. , 2015), exposure to air pollution (Currie et al. , 2015) (Chay & Greenstone, 2005), crime risk (Linden & Rockoff, 2008) and traffic (Tang , 2021). We focus on the restoration of an abandoned canal the Droitwich Canal in the West Midlands of England which was closed in 1939. By the early 2000s the canal was overgrown, drained of water, non-navigable or completely destroyed. The canal underwent a major restoration in 2007 and was re-opened in 2011.
The restoration provided an avenue for recreation activities such as boat navigation, improved the environment and provided a habitat for aquatic life. In our study, we compare price changes for properties close to the canal after the canal is restored (treatment group) with price changes for comparable properties that are unaffected by canal restoration (control group). The assumption behind this method is that prices would have evolved in the treatment group close to the Droitwich canals in much the same way as in the control group, if the Droitwich canals had not been restored.
The validity of this assumption depends on the comparability between properties in the treatment and control group. Hence, we select two alternative plausible control groups: (1) properties close but slightly further away from the Droitwich canal; and (2) properties near to an existing neighbouring canal the Worcester and Birmingham canal that remain in continuous use and are not affected by the restoration over this period. This difference-in-differences analysis gives us by guest on November 13, 2022. Copyright 2021 6 results from a more robust causal estimation strategy, albeit on a specific case, which we can triangulate with the cross-sectional estimates for all England and Wales.
To preview our findings, analysis of the entire canal system in England and Wales suggests that proximity to canals increases house prices, although the effect is highly localised. Properties within 100 metres of a canal have a price premium of around 5% relative to those beyond 1km (estimated on the whole 2002-2016 study period). There is no impact on prices in the 100m-1km range. The local geographical extent of this effect suggests it is associated with canal-side properties and others which have immediate access or views of these waterways. The effect is bigger around 10% -in dense urban areas. Similar effects are detected for the re-opening of the Droitwich canals.
The restoration leads to a 10% increase in values for properties within 100m of the restored canals.
We also investigate the association between canal proximity, and the share of new-build homes sold, as a proxy for housing construction. We observe that the proportion of new-build sales is 5.9 percentage points higher within 100m of a canal compared to further away, representing a 75% relative increase.
Additional analysis reveals that there was a step change in the valuation of this amenity at the time of the recession in 2008, which persisted up to the end of our study period. The premium for living with 100m of a canal halved from around £520 per year pre-recession to £260 post-recession and the proportion of new-build transactions fell accordingly. Back-of-the-envelope calculations indicate the land value uplift from the canal network was around £0.8-£0.9 billion in 2016.
The remainder of the paper is as follows. Next, Section 2 describes the data and the methodology in detail. Section 3 presents and discusses the results of the analysis and Section 4 concludes.

Regression Specifications for National Analysis
The underlying reasons for wanting to live near to a canal are that doing so reduces the time and cost of travelling to them and/or because it means a resident has a direct view of a canal or a canal frontage ii . We therefore use indicators of distance from a property to its nearest canal as our key variables of interest and estimate to what extent prices are lower or higher at different distances (holding other factors constant). Distances are based on full postcode, where a postcode typically contains around 17 dwellings.
We first estimate the house price premium associated with proximity to canals with a standard hedonic property price regression, estimated using ordinary least squares (OLS), that takes the following form: where , the dependent variable, is the natural logarithm of the price of property i located in neighbourhood j and sold at time t. There are, potentially, many unobserved confounding factors that vary with distance to a canal and affect the price directly. These include the physical characteristics of the housing, amenities like distance to employment or proximity to public transportation nodes. For example, canals in urban areas are usually found in old industrial areas, and properties near these areas are typically older and smaller. Industrial buildings in the neighbourhood could also be a dis-amenity and affect housing values directly. Canals also generally follow natural lines along valleys that avoid gradients, so their routes may be topographically distinct from other areas in the country.
To control for these kinds of confounding factors we take the following steps. Firstly, we focus the analysis on a buffer zone close to canals, restricting the sample to properties within 1500m of them. The coefficients on the distance bands in equation 1 therefore estimate the price premium relative to properties in the 1000-1500m band. In this way we compare properties that are all along canal routes, but enjoy different levels of access and exposure to their environmental benefits. We further control for a rich set of fixed and time-varying housing, geographic and location characteristics denoted by ′ . These characteristics are recorded at either the property, postcode Note, the structure of equation 1 also helps us establish that the price premium estimates we obtain are causal, in that we would theoretically expect to see a distance decay profile in the estimates, with the price premium decreasing as distance and travel cost to the canal increases.
In additional analyses, we look at heterogeneity in the price premium across various observable dimensions related to the geography of the location. We also explore the way the price premium and willingness to pay for canal proximity has changed over the years in our study period.
We discuss these regressions in the section where we present the results.
Demand for housing near canals could lead to increases in the supply from developers. We investigate this issue by examining whether a transaction is more likely to be a new-build sale if it is closer to a canal. To implement this analysis, we replace the dependent variable in equation 1 with an indicator of whether a sale is a new build.

Difference-in-Differences Estimation
Even with the inclusion of a large set of controls and geographic fixed effects, and the restriction to property sales around canals, the method described above might fail to control for fixed confounding factors at a very localised geographical level, below MSOA or LSOA. A more robust approach would be to estimate from changes over time in canal access and property prices at the postcode level, thus partialing out fixed confounding factors at this level (given that distance is calculated at postcode level). The limiting factor in applying this approach to the evaluation of the environmental benefits of canals is, of course, that accessibility to canals and their environmental benefits is rarely changing.
One exception is where there have been substantial canal restoration projects, bringing disused, buried and derelict canals back into use, and restoring their value as environmental and Thi project will bring the canals into navigable use and will create a unique 21 mile cruising ring linking Droitwich Spa to Worcester, which can be completed in a weekend by boat. The project is not solely about navigation as it includes many works to enhance the canal corridors as a recreational and environmental resource for local people as well as visitors to the area. Canal restoration will provide a stimulus to the local economy by encouraging tourism-related businesses and will provide many benefits to the local community. It is intended that the vision will be delivered through a series of objectives including: To restore the canals to good navigable condition; To use the canals as a The project was evidently very ambitious in its environmental and recreational aims, and so potentially provides a useful experiment for estimating the value of these benefits to local homeowners. To implement this idea in a difference-in-differences design we need to define treatment and control groups. The control group needs to be carefully chosen such that it is likely to have followed the same counterfactual trends in outcomes as the treatment group would have done in the absence of the policy (the "parallel trends" assumption). As candidate control groups, we select properties that: (1) are between 1000 and 1500 meters from the Droitwich Canal -as in our national analysis; or (2) up to 1500m from the Worcester and Birmingham Canal, a canal in the same geographical region and local economy as the Droitwich Canal, and connected to them, but which was in continuous use over the period and experienced no restoration project or consequent by guest on November 13, 2022. Copyright 2021 12 improvement in environmental quality. Figure 1

presents a map of the Droitwich Canal and
Worcester and Birmingham Canal overlaid on a satellite photograph, making the general layout and similarity in the landscape crossed by each canal clear.
Another key element in our setup is the definition of the "treatment" date when the benefits from the restoration of the canals to start to materialise which we refer to as the post-restoration date.
The project extended over several years from the mid-2000s and there was some restoration activity well before that. There are two plausible choices of this post-restoration date in relation to the major restoration scheme that started in 2007. One date is the submission of planning applications around May 2007. A second is the completion and opening around September 2011. We explore the impacts using one or the other, by estimating the following regression specification: where represents a set of distance band indicators at 100 metres interval and up to 1km (K= 1,2, ,10 ) computed based on the Euclidean distance of property i from the nearest canal, which could be Droitwich Canal or Worcester and Birmingham Canal. is an indicator variable denoting whether property i is within k distance band from Droitwich Canal (e.g if k = 1, property i is within 100 meters from Droitwich Canal). Postt is an indicator denoting whether property i is sold after Droitwich Canal is restored. , … . . capture the price changes with distance from the Worcester and Birmingham Canal after restoration of the Droitwich canals. The key parameters of interest, , … … . , are difference-in-difference estimates that measure the additional property price changes after canal restoration across different distance bands from Droitwich Canal. We constrain our analysis to properties not more than 1500m from either Droitwich or Worcester and Birmingham Canal to mitigate the risk of unobserved neighbourhood differences from biasing our estimates. Hence, these difference-in-differences estimates compare the distance decay in the price

Data Sources
The main source of data for the analysis set out above is the Land Registry p ice-paid dataset that provides detailed information on transaction prices and some basic characteristics. This dataset has been linked to information from Energy Performance Certificates (EPC), which are required for all properties bought and sold in England and Wales iv . The EPC data provides a much richer description of the structure of the property. Although the EPC information only dates back to 2008, the information can be used for properties with EPCs, when they were sold in earlier periods concerned that properties closer to canals could be more or less accessible to these transportation modes, given that railroad and canals often follow the same transport corridors. Distance to rivers and distance to green space is taken from the OS Open Rivers and Open Greenspace datasets (Ordnance Survey, 2018a,b). Land use comes from Landcover map Landsat remote sensed data (Rowland, 2017), each postcode assigned the land use at its centroid, and categories aggregated up to 9 major groups, urban, suburban, and a rural land cover types.
Using the location of each sale, we further map each postcode to Census data units, the of data (matched to all years of transaction data). The data sources are set out in Table A1.

Descriptive Statistics
Our  Table A2. Since our analysis compares house prices in places close to canals with prices in places further away, the table splits the information into three groups 0-100m from a canal, between 100 and 1000 metres of a canal, and between 1000 and 1500m of a canal. These summary statistics show that there are differences between properties sold close to canals and those further away on some dimensions, but not others and it is hard to observe systematic patterns.
Evidently, simply looking at mean prices is not very informative. On average, in these unadjusted figures, property prices are slightly higher in the 100m zone than the 100-1000m zone, but both of these zones are slightly cheaper on average than the zone beyond 1500m. The estimated gap between prices in the 100m zone and the 100-1000m depends on how it is measured, around 1% in the simple means, around 5% when based on the average differences in log prices ( Table A3. Here, we report means and standard deviations for the three distance groups related to the Droitwich Canal (<100m, 100-1000m and 1000-1500m) and for the overall sample for the Worcester and Birmingham control group (<1500m from the Worcester and Birmingham canal).
Again there are dissimilarities along some dimensions when we compare these groups. However, the patterns are different from those in the full England and Wales sample and even less systematic.
Properties 100m from the Droitwich canal are marginally smaller than those 100m-1000m away, and considerably smaller than those near the Worcester and Birmingham canal. There is a higher proportion of terraced houses close to the Droitwich Canal than elsewhere and more social renters. In general, statistical tests of the difference between these groups indicate that only a few of the differences are statistically significant. The simple mean price differences are not revealing of any strong patterns. The results of the difference-in-differences analysis using these data are presented below in Section 3.

Regression Estimates for National Analysis
The results from the regression analysis discussed in 2.1.1 are presented in Table 1. Column 1 of 18 4, when we control for LSOA fixed effects and neighbourhood demographics including education, ethnicity, and unemployment it is likely we are over-controlling, and that the estimated price premium is an underestimate. The estimates are also less precisely measured (wider confidence intervals). The reasons for this are firstly that LSOAs are relatively small spatial units, so within each LSOA there is relatively little variation in distance to canals, particularly in dense locations.
Also the problem with controlling for demographic characteristics is that these will respond to the local housing price, because people chose where to live based on the housing costs. Poorer, less educated and ethnic minorities tend to live in lower cost places. This implies that including controls for these demographics may eliminate some of the price effects we intend to estimate. We therefore regard column 3 and 4 as robustness checks, and our preferred estimate is that in column 2. These estimates across distance from canals are plotted in Figure 2a.
How should we interpret the key result from These results do not identify any specific feature of canals that might be attractive. In additional analysis we looked at the effects of specific features locks, aqueducts, wharves, and canalised rivers alongside the basic effects of canal proximity. We found no interesting patterns related to aqueducts or wharves, but there is a significant (at 10% level) price premium associated with canal locks, and an insignificant effect of canalised rivers within 100m, of a similar magnitude to that for canals vi . This pattern for locks is illustrated in the online Appendix Figure A2. There is an additional effect from locks, of around 4.5% within 100m falling to 3% at 200m, although the estimates are only statistically significant at the 10% level. Some of this effect may be driven by the desirability of former lock keepe canal-side cottages, but there may be some heritage value associated with locks in general.
In the next analysis in this section, we look at how the price effects from canal proximity vary by type of location. Here we focus only on the effects of being within 0-100 metres, given the lack of any effects elsewhere. Table 2, column 1 shows the differences by built-up urban and non-urban locations (using the land cover categories described in Section 2). The first row of column 1 indicates that outside urban areas, the price premium for the 0-100m band is 2.7%. This increases by an additional 7% in urban areas, making the total effect in urban areas around 10%. A plausible explanation for this finding is that canals offer particular environmental and recreational benefits in urban areas, where there is limited green space available, and canal-side locations may be particularly coveted. Urban in this land cover data refers to the densest parts of cities.
Column 2 repeats the analysis for suburban and urban areas, which represent over 95% of the sample. Here we can see that all of the basic premium for canal proximity is driven by urban and suburban locations, and the effect in rural places (given by the first row of column 2) is by guest on November 13, 2022. Copyright 2021 20 insignificantly negative. The implied premium for living within 100m in urban and suburban areas in these estimates is 5.9% (this is slightly higher than in Table 1, because here we are comparing 0-100m, with 100-1500m). We also double checked for effects at higher distance bands in the urban/suburban sample but found none. Column 3-4 look at differences by whether a property is close to other rivers or green space which might provide alternative recreational and environmental services, but we find no evidence that this matters in general in the national sample, even if it matters to urban populations as evidenced by column 1.
It is useful to translate the percentage premium on house prices into monetary equivalents, which represent willingness to pay for canal-side amenities i.e. how much households are willing to give up on other expenditure in order to enjoy homes close to canals. Some care is needed in doing this, as we have estimated an average percentage premium over the whole period, but average house From then on it fell considerably, the obvious explanation being a shift in the housing market following the great recession in 2008. It is well known that the character of the housing market has changed since then, with much lower transaction volumes.
In Table 3  interactions with year indicators, analogous to Figure 2b. The pattern is startling: the probability of a transaction being a new build was around 8% higher closer to canals at the beginning of the period, but started to fall rapidly around the time of the 2008 recession, with new build transactions becoming less prevalent close to canals than further away by 2011. Coupled with the evidence on the fall in the price premium over this period, these results suggest a drop in new build transactions as demand and willingness to pay for canal-side locations fell. Other work has shown a fall in willingness to pay for environmental amenities during the recession (Cho et al. , 2011), but we are not aware of other work that has found that these effects might be long lasting.
One concern, given these results on the sales of new build properties close to canals is the possibility that part of the price premium we observed for canal-side locations could be due to new builds, if new builds are more prevalent in canal side locations and if new builds command higher prices. Although our main price regressions controlled for whether or not a transaction is a new build, we investigated this question in more detail by re-estimating the property price regressions: on second hand homes only; with interactions between new-build and canal-proximity dummies; with interactions between new-build and canal-proximity dummies, alongside interactions between urban and canal-proximity dummies, and alongside urban new-build canal-proximity dummies.
The results are shown in Table A4 in the online Appendix. Although we find that there are indeed high premia for new properties in canal-side locations, and even higher premia for new build properties in urban areas in canal-side locations, these are not the primary drivers of the price premium we observe in our main estimates.

Difference-in-Differences Estimates
In this section, we report the results of the difference-in-differences analysis of the Droitwich Canal restoration described in Section 2. As discussed in that section, these results relate to the impact that  The plots in Figure 4a and 4c bear a similarity to those from the national estimates in Table   1, although the methods used to estimate them are substantially different. Here we are estimating only from the changes in prices over time near the Droitwich Canal around the time of the start of the major restoration, compared to the changes over time occurring over the same time in the control group. The effect of the restoration within 0-100m is large before opening, at around 11.9%, although there is a marked decline after opening. Taken together the overall impact reported in Figure 4c is around 7.6%, which is slightly larger than the 5% found on the national cross-sectional analysis in Table 1, although given the wider confidence intervals the figures are statistically similar.
These patterns of distance decay in these estimates are not so clear cut, with some evidence of price by guest on November 13, 2022. Copyright 2021 24 uplift in 400-1000m bands. It is possible that the effects are spuriously related to confounding factors specific to the Droitwich area compared to the control Worcester area. Nevertheless, the sharp distance decay between 0-100m and the rest provides some assurance that the 0-100m effect can be ea ed a a ca al impac of he canal e o a ion on immediately proximate property prices.
The estimates from this difference-in-differences evaluation are less precise than those from the cross-sectional analysis, and are based on a single case study area and much smaller sample.
There are risks in looking at a single case like this, in that the estimates may be influenced by local million. This figure of course ignores the homes that have not yet sold, the value uplift to land that has yet to be developed, and the value ignores any benefits not captured in the housing market.

Conclusions
Canals potentially provide a desirable recreational and environmental amenity. In this paper, we estimate the monetary-equivalent value of this amenity to local residents using house prices. The revealed preference framework adopted in this study is a standard approach to valuing non-market corresponding to annual monetary willingness to pay of around £520 pre-recession, and £260 postrecession, in 2016 prices. We find no price premium for living close to a canal but beyond 100m, which suggests that the effect is driven predominantly by canal-side properties, and others with a direct outlook on the canals or immediate access. We further observe a higher proportion of newbuild sales within 100m of canals relative to elsewhere over the period -a 5.9% increase on a 7.8% baseline, so around 75% higher -suggesting considerable response in construction to this demand for canal-side homes. However, the rate of new build sales fell dramatically post-recession, in tandem with the fall in the price premium. A unique application of a difference-in-differences evaluation methodology to the restoration and environmental rehabilitation of the Droitwich Canal in the West Midlands supports the key findings on prices.
As an interesting, if very imprecise exercise, we calculate the potential implied land and property value uplift from the canal network. The length of the network covered in this analysis is 3500km. The price effects extend over 100m either side of the canal, so the affected area is 0.2 3500km = 700km 2 , which is just under half the area of Greater London. Though we do not have the exact figure in our data, around 10% of the land of England is urban/suburban and so developed or hypothetically developable, so the price uplift from canals would affect about 70km 2 , or 70 million      Table  A1 for the exact list of variables included for each set of controls. Absolute price change is calculated by multiplying the average transacted prices on the estimated premium (main effect plus interaction effect) for properties within 100 meters from canal. Average transacted price for regression sample is £234,908. Standard errors are clustered at LSOA level. *,**,*** denotes significance level at 10%, 5% and 1% respectively. ii The values that can be elicited through house prices are therefore what environmental economists refer to as use value, as opposed to, say, the satisfaction one might get from just knowing that such a resource exists without ever intending to visit or use it.
v For more details, refer to https://data.gov.uk/dataset/660ab8be-2912-4ef5-a8a9-7ed3111e34d1/canal-centre-line vi The coefficients on our control variables indicate that there is also a premium of a similar magnitude for living near other natural rivers that extends over a wider range of distance, but again these are not statistically significant, and not the primary focus of this analysis.
vii These urban land cover figures come from NEA (2011).
viii This value is relative to other places, so is not necessarily an addition to the total value of the land or housing stock in England and Wales.