Do Banks Generate

Financial Market Integration?

Liam Brunt1 & Edmund Cannon2

Abstract3

Using a large panel of weekly wheat prices, we infer the annual rate of return on capital in each county in England and Wales in the period 1770-1820. Throughout this period markets were efficient in the sense that weekly returns were serially uncorrelated. We show that the interest rate differential between London and each county can be explained by the density of bank coverage in that county. The explosion in provincial banking in England and Wales during the industrial revolution significantly reduced regional differentials in interest rates. This is direct evidence that the depth of financial intermediation determines the degree of capital market integration.

Keywords: banks, financial integration, industrial revolution.

JEL Classification: O16, N13, G21.

1 St. John’s College, Oxford OX1 3JP. . [email protected] 2 University of Bristol, Department of Economics, 8 Woodland Road, Bristol BS8 1TN. United Kingdom. [email protected] 3 The authors gratefully acknowledge funding by ESRC Research Grant No. R000220371 and a Royal Economic Society Small Grant to help buy some additional software. The authors would also like to thank Ludivine Jeandupeux for research assistance; Rob Brewer, Anna Chernova, Becca Fell, Saranna Fordyce, Dave Lyne, Olivia Milburn, Hannah Shaw, Derick Shore, Liz Washbrook and Alun Williams for helping with obtaining, entering or checking the data set; Colin Knowles for computing assistance; the Bristol City Library for allowing us to borrow copies of the London Gazette not normally available for loan; and Cliff Attfield, Giam Pietro Cipriani, Suro Sahay, Lucy White for other assistance and helpful comments.

1 Introduction. Britain was distinguished from other economies during the Industrial Revolution by the sophistication of her financial markets (Neal, 1990). This is often invoked as an important cause of her industrial success in the eighteenth and nineteenth centuries (see, for example, Cameron, 1967). The importance of efficient capital markets, the difficulty of achieving them and their rôle in generating economic growth has also received considerable attention from development economists (Ikhide, 1996; Sial & Carter, 1996; MacKinnon, 1976; King & Levine, 1993). The growth of commercial banking in Britain was truly spectacular, rising from 120 banks in 1784 to 660 banks in 1814 (Pressnell, 1956). So if banking does indeed have a positive effect on the real economy then we should certainly be able to observe it in action in the British Industrial Revolution. In fact, there is very little direct evidence linking financial market integration to changes in the cost of capital during the Industrial Revolution; and a link between the cost of capital and the rate of investment during the Industrial Revolution has never been established empirically. Without these two links in the chain it is difficult to estimate the effect of financial market integration on British economic growth. The purpose of this paper is to quantify the effect of commercial banking on financial market integration in Britain during the Industrial Revolution. The fundamental problem in analysing financial markets in this period is that we have very little information on the cost of capital. Hence it is difficult to estimate the effect of any changes that might take place on the demand or supply side of the market, or any change in the nature of financial intermediation. The recent analysis by Buchinsky and Polack was based on two series of deeds on property transactions, one for Yorkshire and the other for Middlesex, from which they inferred regional building, presumed to be determined partly by the cost of capital. Whilst their results are suggestive, one would not want to place too much weight on them alone. Our departure point is to estimate the rate of return on capital in each county. We can do this by looking at the appreciation of a real asset through the year. The only asset that is sufficiently well documented through the period is grain. Bearing in mind that agriculture was the largest sector in the British economy until 1840 (Crafts, 1986) - and that grain was the single most important output - the rate of return on holding grain is probably the single best indicator of the cost of capital in the economy. McCloskey & Nash (1984) and Taub (1987) show how the seasonal variation in grain prices is related to the interest rate: grain is an asset, and in equilibrium the holders of grain must be compensated for storage and interest costs. We use the

2 appreciation of grain prices through the harvest year to estimate the rate of interest prevailing in each county from 1770 onwards. We have previously used this technique successfully on data from earlier periods (Brunt & Cannon, 1999). Our estimates are based on a large panel of weekly grain prices collected by the British Government between 1770 and 1820; this enables us to estimate year-specific county-specific rates of return on capital. We compare the movement of our interest rate series over time to that of the Consol rate, and find a positive relationship. We then explain the geographical and temporal variation in interest rates with reference to the spread of banks outside London (the so-called ‘country banks’). At this time the Bank of England enjoyed great privileges in the commercial banking market. As a result, other banks were restricted to being partnerships (as opposed to joint stock companies) with a maximum of six partners all of whom had unlimited liability. Hence the size of individual banks was greatly circumscribed and the geographical reach of each bank was inevitably very limited. We take advantage of this fact by using the number of banks as a measure of the availability of banking services within each geographical area. There is county-level data available for country banks from 1800 onwards. We show that the density of country banks has a significant effect on narrowing the differential between the rates of return on grain traded in London and each county outside London. The remainder of the paper is organised as follows. Section 1 discusses the merits and problems of using grain prices to infer the cost of capital. Section 2 discusses our data set in more detail and illustrates the seasonal pattern of prices more closely. Section 3 begins with a discussion of market efficiency in the sense used in the finance literature - namely that returns should be serially uncorrelated and prices follow a martingale process (weak market efficiency). This is important because it influences our interpretation of the price data. We then estimate the gross rate of return on grain for each county-year. Section 4 outlines the institutional structure of British banking in the period 1770 to 1820, which informs our model of financial market integration. Section 5 uses the county-level interest rate estimates to analyse the level of financial market integration, and shows how this was influenced by the spread of banks. Section 6 concludes.

1. Inferring the cost of capital from grain prices. There are several conceptual and practical difficulties with inferring the cost of capital from the appreciation of grain prices. This has

3 generated considerable scepticism amongst economic historians about the value of the approach. In this section we address these issues and explain why this is the most practical and effective method of estimating the cost of capital. In the following section we discuss the technical details of our estimation procedure and implement it. There are two lines of argument which we can adopt to defend grain prices as a data source for inferring the cost of capital. First, the grain price approach has the advantage that it is based upon a considerable body of data and provides high frequency data which is at least available on a consistent basis and is as good quality as anything else available. Second, there is qualitative evidence that markets behaved in the way assumed by the grain price approach and that thus it gives us an accurate and representative estimate of the cost of capital. Hence it is perfectly reasonable to use grain prices, whether or not there are other sources. We now consider each of these arguments in turn.

1.1 Options for measuring the cost of capital. The usual approach to measuring the cost of capital is to look at the rate of return on a riskless asset or, if that is unavailable, to use widely traded assets whose risk can be quantified easily. The usual asset in these circumstances would be short-dated treasury bills or government bonds, since such government-backed securities are free of firm- or individual-specific risks. However, if we want to measure interest rate variation within a country that has poorly integrated financial markets then our task becomes more difficult, since government bonds are generally issued and traded only in the capital city. The best that we can hope to do is to find an asset which is widely traded and is equally risky in each place; this would then allow us to measure genuine variations in the cost of capital, rather than variations in risk across the country. The second best would be to look at assets in each place which have different but known levels of risk, which would allow us, in principle, to infer risk- adjusted rates of return. When dealing with historical economies, there are three data sources that offer a reasonable chance of allowing us to estimate the local cost of (riskless) capital. The first source is bank records. James (1978) has used this source for the in the late nineteenth century. Unfortunately, it is exceedingly difficult to get good data for earlier time periods and other countries. Moreover, there is a huge limitation with this approach. If we want to examine the effect of banks entering a local capital market ab initio, then we need

4 to observe the cost of capital both before and after the banks exist. But it is clearly impossible to use the records of non-existent banks to measure the local cost of capital, and hence we can never observe the cost of capital prior to the advent of banks. This is highly likely to be a binding constraint; when we are dealing with any economy of the nineteenth century (and many developing economies today), particular localities will have no banks during at least some of the period of analysis. The second source of data on the cost of capital is mortgages on land, which is both widely traded and relatively low risk. Allen (1988) has used this source for England in the early modern period. Unfortunately, it is difficult to get a large and geographically dispersed sample of mortgages. Moreover, there may be severe sample selection problems. For example, we have often been left mortgage documents from the largest landowners, who may have had peculiar risk characteristics. Also, the large landowners may well have had exceptionally easy access to the largest financial markets. So the mortgage rate recorded for a piece of land in the north of England may actually reflect the cost of capital in London (where the landowner raised his loan), rather than the cost of capital in the north of England. The third data source on the cost of capital is grain prices. The grain production process results in grain being harvested once per annum (generally in August in England) and then stored through the next 12 months. Whilst it was stored, the grain had to appreciate enough to offset the physical cost of storage, the physical losses in storage, and the cost of the capital invested in the purchase of the grain. Otherwise, rational economic agents would not have stored the grain, instead finding some other investment opportunity for their savings. Following the work of McCloskey & Nash (1984) and Brunt & Cannon (1999), we can use this price appreciation to estimate the local cost of capital. Grain prices offer several advantages over other sources. First, before the Second World War every locality produced and traded grain for local consumption. Second, governments keenly monitored the price of grain in every locality because they were worried about the possibility of excessive price rises causing social unrest. This has left us with a huge amount of price data, usually with weekly observations (or sometimes even higher frequency). Third, holding grain was roughly equally risky in every locality, so we do not have to worry about the risk premium when looking at regional variations. Given the relative merits of the three sources for estimating the local cost of capital (bank loans, mortgages and grain prices), the grain price approach is a clear winner. The potential

5 pitfalls in using grain data are no greater than the potential pitfalls with the other sources; and grain prices are the only source which are likely to give a wide and consistent spatial and temporal coverage.

1.2 Accuracy and representativeness of cost of capital estimates based on grain prices. If our estimates of the cost of capital are to be both accurate and representative of the wider economy, then certain conditions need to be met at the micro level and the macro level. First, at the micro level the appreciation of wheat stocks can be interpreted as a return on investment – that is, a rate of interest – only if the grain price is determined by the activities of homo economicus. Economic agents must be (at least approximately) rational, far-sighted, self- interested and operating in a market economy. If agents are pursuing some other kind of behaviour or maximand then grain prices will clearly not exhibit a return on investment. Komlos & Landes (1991) have argued strongly that agents in the grain market cannot be assumed to be homo economicus. This criticism focussed on the medieval economy (rather than the eighteenth century) but many economic historians will have similar reservations about farmers in the later period and we need to address their fears. Second, suppose that we can answer the first question in the affirmative and treat the appreciation of grain prices as a rate of interest. Can we then go further and interpret this return on investment as the rate of interest? This depends on the extent to which product and financial markets were integrated; and the extent to which agriculture, industry and services were integrated. We begin by noting that the conditions required for the appreciation of grain stocks to reflect a rate of interest are actually quite mild. We do not require everyone in the economy to be a rational, far-sighted, self-interested optimising agent. It could be the case that many people held grain irrespective of whether they were expecting to receive a return on the money that they had invested. But as long as agents at the margin required a market rate of return, then the price of grain would exhibit a cycle that ensured a rate of return for everyone. That is to say, the existence of people who are willing to arbitrage between grain markets and other markets (especially financial markets) is sufficient to ensure that grain holdings earn the market rate of return. However, it would obviously strengthen our case if we could show that most or all participants in the grain market were likely to seek a market return on their stocks of grain. So we now consider each type of agent operating in the grain market. We show that in each case

6 they acted as rational economic agents and were well-integrated with financial markets, arbitraging between the grain and financial markets as necessary. Let us start with the farmer. English agriculture in the late eighteenth century was very commercialised. The standard pattern was for large landowners to rent out substantial farms (say, 200 acres) to tenant farmers on long term contracts (usually lasting 10 years and sometimes 20 years). These tenant farmers provided all the movable capital (animals, tools, seed, etc.) and hired workers as required from an active labour market (both on annual contracts and on a daily basis). The tenant farmers were geographically mobile and often moved from one locality to another between tenancies. They were used to evaluating investment opportunities and calculating rates of return; and generally they were not capital constrained. In fact, agricultural profits were very high in the late eighteenth century and the investment funds for industry came from the agricultural sector. It is therefore no surprise to find in 1796 that farmers were well aware of the wheat price cycle, and they arranged their grain disposals accordingly. [T]here are a set of wealthy farmers who have it in their power to retain a part of their growth in those natural and best of granaries, their ricks. Was it otherwise, as the Corn Laws now stand, we might often, even with a most plentiful harvest, be in the utmost danger of famine. The argument made use of is, that the little farmer is, through necessity, obliged to thresh out his corn and bring it to market; but that the opulent man will not produce his, until it comes to a certain price. (Arbuthnot, 1796, vol.27, pp.21-22)

We have evidence also from the testimony of farmers and merchants appearing in the Parliamentary enquiry of 1828 (British Parliamentary Papers, 1928, vol.18, pp.284-9). The witnesses to the committee discuss both ‘normal’ trade conditions and those pertaining during the agricultural depression of the 1820s. We can see that farmers commonly financed the storage of grain through bank lending, in anticipation of higher prices later in the year. During the war, the landlords easily raised money at the banks on discount, and consequently were not under the necessity of opposing the speculations of the farmers; since however, the failure of so many banks in the south of Ireland, the landlords, generally, have been unable to continue this indulgence. We find accordingly, by the notes of the different markets, that the delivery of crops is every where, not only quite unreserved, but much earlier than during the war.

And again, The want of money has obliged them generally to bring their Corn to market as early as possible of late years, and but few of the more wealthy have seen sufficient prospect of advance to induce them to hold [Corn].

7

This account is particularly interesting because we can see how higher interest rates in the banking sector are reflected in higher rates of return on holding grain. The tightening of credit rates makes it difficult for farmers to borrow money to finance the holding of grain; so they bring their crop to market earlier, and this drives down the price of grain early in the season. This generates a sharper increase in grain prices over the year, reflecting the higher interest rate pertaining in the banking sector. Grain factors, millers and mealmen (that is, flour merchants) also sought a market rate of return on their stocks; and they were also commonly financed through bank loans. Hence in the depression of 1828 we find that: The capital employed in the Corn trade has been less of late years from less accommodation by country bankers and others…

The involvement of English banks in the grain trade was apparently very extensive, and the state of the grain trade was consequently sensitive to conditions in the capital market: The capital employed in the Corn trade in the South of Ireland is very much supplied from England, the corn being bought on commission. The failures of the banks have certainly lessened the capital in the Corn trade in Ireland, by lessening the accommodation afforded to persons who have no capital of their own.

Two further points are worth emphasising. First, agents who were intimately involved in the grain trade were perfectly willing to withdraw their capital from the grain trade and redirect it to more remunerative avenues. This is important because if the rate of return on capital is to be equalised between the grain market and the banking sector then it is essential that capital can flow out of the grain market as well as into it. The capital varies according to the price of Corn. Many millers have laid out a part of their capital in the funds [government securities]…

Similarly, Most of the capital employed by merchants or middle men in former years has been lost, and the remainder withdrawn from the trade, there being now no capital employed in the trade in Corn.

Second, we also find that the holding of grain purely as an investment activity was not limited merely to farmers and grain factors. It was also seen as a reasonable target for other investors seeking a return on their capital:

8 …but many capitalists and merchants unconnected with the [grain] trade, were wont to speculate in Grain, through the factors…

Finally, if we move further forward in time to the 1870s then we find Bagehot (1873) describing how the bill brokers directed funds to flow into and out of the grain market as necessary. Their bill cases [i.e. portfolios] as a rule are full of the bills drawn in the most profitable trades, and ceteris paribus and in comparison empty of those drawn in the less profitable. If the iron trade ceases to be as profitable as usual, less iron is sold; the fewer the sales the fewer the bills; and in consequence the number of iron bills in Lombard Street is diminished. On the other hand, if in consequence of a bad harvest the corn trade becomes on a sudden profitable, immediately ‘corn bills’ are created in great numbers, and if good are discounted in Lombard Street. Thus English capital runs as surely and instantly where it is most wanted, and where there is most to be made of it, as water runs to find its level.

In order for the rate of return in the grain market to reflect the rate of return in the financial market it is not necessary for there to be direct links between the two sectors (i.e. banks investing in the grain market). But the existence of direct links obviously greatly strengthens our case. The evidence presented above demonstrates that capital could easily flow into and out of the grain market and that there were strong direct links between the grain market and banks. First, grain market capital came from individual agents who were able to trade directly in both grain markets and financial markets (i.e. their investment in grain was not mediated through putting their money into a bank). We can see this process occurring when individuals bought grain stocks through grain factors. Second, agents who were themselves involved in the grain trade held large grain stocks if they expected a high rate of return; but they were also willing to withdraw their capital from the trade and find alternative investments if there were higher expected returns outside the grain trade. This suggests strongly that capital could flow out of the grain trade just as easily as it could flow into the trade. Third, financial institutions were involved in financing both the grain trade and other trades – so they were willing and able to arbitrage between the two markets. We can see this process occurring when country banks financed farmers and traders who wanted to hold grain stocks. All this evidence demonstrates that grain markets and financial markets were linked directly by a host of rational economic agents, drawn from inside and outside the agricultural sector. Therefore it is reasonable to interpret the

9 appreciation of grain stocks as a rate of interest, and we would expect this to reflect the rate of interest.

1.3 The nature of capital and the rate of return in the British economy. We have argued that we can estimate the rate of return on capital using grain prices because there were strong links between the grain market and other markets (and hence the returns should move together). But this really understates our case. Suppose that markets were segmented and rates of return therefore differed from one to another (returns in the shipping industry differed from iron and steel, which differed from agriculture, which differed from the insurance industry, and so on). Moreover, suppose that markets were geographically segmented and rates of return differed from place to place (returns on shipping were not the same in London and Liverpool, and so on). Then if we wanted to measure geographical variations in the cost of capital, what could we reasonably take to be the rate of return in each locality? It would be no use comparing coal in Northumberland to grain in Norfolk because the coal and grain markets would have different risk characteristics. And it would be no use trying to compare coal in the two places because Norfolk does not produce any coal. In modern economies it might be natural to take the return on equities, or the rate of interest on housing loans (bearing in mind that we cannot use government debt because we are interested in regional variation). But both of these markets were small in the British economy until the late nineteenth century. We need a market that attracted a large volume of capital everywhere. Which market attracted the most capital in the eighteenth and early nineteenth centuries? It was, of course, the grain market. Grain had to be stored throughout the year. If we assume that the rate of consumption was constant from one harvest to the next, then grain stocks would decline linearly. So on average the amount of grain in storage would be equal to half of the total quantity harvested. Since we know the size of the harvest and the price of grain, we can thus calculate the average value of circulating capital which was invested in grain stocks on any day of the year. We made this calculation for 1760 and 1860 and then compared the value of circulating capital invested in grain stocks to that invested in the non-farm sector using the data in Feinstein (1978). The results are reported in Table 1 below:

10 Table 1. Circulating Capital in the British Economy, 1760 and 1860 (£m at 1861 prices). 1760 1860 Total for the non-farm sector 40 210 Total for the farm sector 140 240 Grain stocks 48 61

In 1760 the value of circulating capital invested in grain stocks was greater than that invested in the whole of the non-farm sector; even as late as 1860, the capital invested in grain was around one third of the entire non-farm sector. Any particular industry (even leading sectors such as cotton or steel) would have been easily out-classed by the grain market, in terms of circulating capital. Moreover, the leading sectors were heavily regionally concentrated in Lancashire and Yorkshire, whereas the grain market was important in every locality. So as a barometer of the rate of return on capital across the country, the grain market is a far-superior instrument. The immense influence of agriculture on the prosperity of the British industrial sector was being emphasised by Bagehot as late as the 1870s: But every such industry is liable to grave fluctuations, and the most important – the provision-industries – to the gravest and the suddenest. They are dependent on the seasons. A single bad harvest diffused over the world, a succession of two or three bad harvests, even in England only, will raise the price of corn exceedingly, and will keep it high. And a great and protracted rise in the price of corn will at once destroy all the real part of the unusual prosperity of the previous good times. It will change the full working of the industrial machine into an imperfect working; it will make the produce of the machine less than usual instead of more than usual; instead of there being more than the average of general dividend to be distributed between the producers, there will immediately be less than average.

So if we are trying to estimate the interest rate in the British economy during the Industrial Revolution, then it makes most sense to take the grain market as a benchmark case. The grain market absorbed more capital than any other market or sector; and the rate of return that it generated is the one by which all others may be judged.

2. Discussion of the data. Our analysis is based on weekly price data for wheat collected by the Receiver of Corn Returns and published in the London Gazette for each county in England and two regions in Wales: this is fully discussed in Brunt & Cannon (2001).

11 Weekly price data is available for the 40 English counties,4 London,5 North Wales and South Wales – resulting in 43 price series for each grain. For simplicity, we refer to these 43 geographical areas as “counties”. We analyse each of these series from 10 November 1770 (when the data begin) to 30 September 1820. The Welsh data published in the London Gazette are disaggregated further after 10 November 1790, at which point the data are reported for each of the twelve Welsh counties individually. We have simply averaged the six Northern and six Southern Welsh counties respectively to extend the two series through to 1820. There were inadequate data for analysis for Scotland. Data for the period after 30 September 1820 are also available, but some of the inland counties are unrepresented until a further change in the data set in 1828, apart from some other changes in the lists of , so 1820 forms a natural break-point. The county prices are weighted averages of prices in several market towns in each county (where the volume of trade was used as weights). Suffice it to say here that the number of towns monitored in each English county was almost always three. London was obviously a singleton, and the tiny county of Rutland had only two monitored towns; some of the counties have more towns (for example, Norfolk has twelve). By law, the average price for a county could be reported in the London Gazette if and only if returns had been received from at least two thirds of the towns in that county in that particular week. Some of the price series have missing observations, but the scale of this problem is minimal. For the years 1770 to 1820 we do not know why the prices are missing. For the period after 1820 we know that most missing price observations occur because there was no trade in that particular product in that particular week (so there was simply no market price). Over the entire period, the number of missing observations for wheat is very small, and is zero for some counties. The worst offender is Herefordshire where there are 62 missing observations out of 2604 weeks (so data is available for 97.6% of all potential observations). Since we do not have quantity data to correspond to these price data, we cannot draw strong conclusions about the regional pattern of trade or production, but the pattern of missing

4 Monmouth at this time was officially an English county, although normally considered part of Wales. In our analysis of banks we have tended to follow Shannon and merge Monmouth with South Wales. 5 London data were not published after 1793. We have managed to collect weekly wheat price data for the London market from private newspapers for the period up to 1815. Thereafter no more data appear to be available (some newspapers report a price for 1817 to 1820 for London, but it turns out to be identical to the London Gazette national average price).

12 price observations suggests that wheat was widely grown in all areas, even if the climate were unsuitable. This is what we would expect, given the high transport cost of moving wheat. Figure 1 illustrates the time series price of wheat over the period 1770-1820. The inner line represents the mean average price of England and Wales: the two outer lines represent the maximum and minimum price of the 43 counties in each week. This graph illustrates that prices in different regions moved closely together. This fact alone does not allow us to infer immediately that markets were highly integrated. It would be quite possible for the random shocks affecting prices to be highly correlated across counties (notably the weather), resulting in similar prices in markets that were only very weakly linked.

Figure 1: Weekly Wheat Prices Mean, Min and Max (pence per Qtr)

350 300 250 200 150 100 50 0

1770 1773 1776 1779 1782 1785 1788 1791 1794 1797 1800 1803 1806 1808 1811 1814 1817 1820

A further feature of the data is the tendency for the underlying levels and year-on-year variability of prices to be constant until about 1794. After this time there is both a secular increase in prices and much greater variability, due to the problems of the Napoleonic Wars. Ideally we should approach this problem by deflating the price series to obtain the real prices of the grains, but the lack of a suitable price deflator makes this impossible. Our analysis of rates of

13 return will thus be confined to nominal rates of return – but since we are using the nominal rates of return in all places, this will not influence our results. We now turn to the seasonal pattern of the grain prices. The simplest way to describe the seasonal pattern is to conduct a regression of the form

52 52 (1) lnPt = a0 + ∑ ai + et subject to ∑ ai = 0 . i=1 i=1 where a0 is a constant and the ai are 52 dummies for each week of the year. Equation (1) can be applied to each of the 43 county series. The result is shown in Figure 2, which displays the average of all these county estimates. From this graph it appears that there is a strong seasonal pattern which is similar in all counties. However, care must be exercised in interpreting the graph, since in most counties the regression is insignificant using conventional standard errors. Chambers & Bailey (1995) argue on this basis that there is no seasonal effect at all (although inconsistently they maintain that there is a significant price fall in September). However, the use of conventional tests in this context is inappropriate. From Figure 1, it can be seen that the most of the temporal variation in prices is not the seasonal effect within years but the combination of the year-on-year effect throughout the sample and the large inflationary component after 1794. Any attempt to conduct tests on the basis of equation (1) would need to account for the immense autocorrelation and heteroskedasticity present in the resulting residuals et. For this reason we present Figure 2 as only illustrative. The graph reveals that the average seasonal pattern of prices is broadly consistent with the McCloskey-Nash characterisation: prices rise for most of the year and then fall at the point of the harvest. The gross rate of return for the period when prices are rising is about 6 per cent, but this is for a period of about 30 weeks: expressed at an annual rate the returns would be 10.5 per cent. In one aspect, however, the seasonal patterns for wheat depart from the McCloskey-Nash pattern: the period of the year when prices are falling is much longer. In the simple McCloskey- Nash hypothesis, price falls are only possible if stocks of grain are relatively low, or if they are unanticipated. Neither of these cases can be relevant on a consistent basis for the period of three to four months immediately after the harvest.

14 Figure 2 Wheat Mean County Seasonal Effect Nov 1770 - Aug 1820

0.05

0.04

0.03

0.02

0.01

0

-0.01 13579111315171921232527293133353739414345474951

-0.02

-0.03

-0.04

It is necessary to have a more sophisticated description of the pattern of production to understand pricing behaviour in this part of the year. Part of the reason for the pattern is that the timing of the harvest was itself highly variable and thus new stocks of grain could become available at any time between late July and early September. More importantly, grain had to be threshed before it was brought to market. Labour costs were highly seasonal, with the highest wages in the period July-September. This is because labour was needed for the urgent tasks of harvesting and then ploughing before the weather deteriorated in the autumn. After this period, the demand for labour fell considerably and only then did farmers allow much of their labour force to be diverted to threshing, most of which took place in November and December. This remained true into the late nineteenth century (see Young, 1770; Chalmers 1868 for discussion). For this reason the falls in the price of wheat during this period do not fully reflect price rises in the underlying asset and we prefer to concentrate on the period from late December, by which time threshing was usually completed. This section has provided a brief description of the data. We have shown that the seasonal pattern of wheat broadly follows the pattern suggested by McCloskey-Nash and hence we shall be able to use these data to estimate the rate of return on capital.

15 3. Econometric Estimates of the Gross Rate of Return. In this section we present further evidence that wheat prices behaved in a way that we should expect for a weakly efficient asset market and we obtain estimates for annual rates of return. In financial time series analysis it is conventional to use a definition of efficiency known as weak market efficiency, where changes in prices must be unpredictable from past information on prices alone. Tests for this sort of efficiency can be based entirely on regressions involving just price data and it is this definition that we consider here. If wheat prices do behave in the way of normal asset markets then we should expect (the logarithm) of prices to follow a random walk. Consider the example regression,

(2) ln Pt = α +ξ1 ln Pt−1 +ξ2 ln Pt−2 +ξ3 ln Pt−3 + ut .

Under the hypothesis of weak market efficiency, we should obtain the results that ξ1 = 1, ξ2 = ξ3

= 0 and ut is serially uncorrelated. The estimate of α would then be the rate of return. In financial time series it is common to model the variance of the residual to be evolving over time (in the simplest case using just an ARCH specification) but our analysis of prices within a year, with at most 40 observations, precludes this approach. Confining ourselves to OLS means that our estimates may be inefficient and conventional standard errors unreliable. In these circumstances, we use Heteroskedastic Consistent Standard Errors and test explicitly for evidence of ARCH in the residuals (although with so few observations the power of this test will be weak). In fact we find little evidence of heteroskedasticity within years. Our more important problem is in deciding the time frame over which to use regression (2). Since we should like to estimate changes in interest rates over time, the natural approach is to treat each year as an individual time series, resulting in substantially different estimates of α for each year, which we may refer to as αy, where the y subscript denotes the year to which we are referring (reserving the t subscript for the week of the year). However, this makes it less straightforward to interpret the tests for ξ2 = ξ3 = 0 under the strict assumption of weak market efficiency. Weak market efficiency assumes that returns are uncorrelated given information on past prices alone, i.e., not including αy. If this parameter is not known, then ∆lnPt-1 provides useful information about αy and we should not expect returns to be uncorrelated. For this reason

16 our tests of market efficiency are not, strictly speaking, tests of weak-market efficiency because they condition upon αy. It should also be noticed that we do not allow for changes in the rate of return within the year, since such changes could not be reliably detected with our data set. We start our tests by using Augmented Dickey-Fuller (ADF) tests using the re- parameterized and extended equation

(3) ∆ ln Pt = α + bt + ()ρ −1 ln Pt−1 + λ1∆ ln Pt−1 + λ2∆ ln Pt−2 + ut :

Under the alternative hypothesis that there is no unit root the data would have to follow a deterministic trend and hence omitting the bt term would bias the test towards failure to reject the null, even if the null were false. We conduct this test on data for each county for each harvest year from 24 December to 8 August, resulting in 50 tests of a unit root for each county (harvest years 1770-71 through to 1819-20). To check the robustness of these tests we varied the start and end dates of the period within year and varied the number of lags in the ADF test. Given the number of tests, we should expect to reject the null hypothesis about 5% of the time, so it is desirable to summarize the information from all of these tests. Since within each county the different years are independent, it is possible to combine the ADF tests using a technique suggested by Fisher (1932) and advocated by Maddala and Wu (1999) in the similar context of evaluating test statistics within panel data. If the independent probability values of the

ADF tests are denoted πi, then the statistic

N (4) − 2∑lnπ i i=1 has a χ2 distribution with 2N degrees of freedom. Maddala & Wu’s (1999) Monte Carlo tests show that the test statistic has almost exactly the right size when N = 50 and T = 25, which corresponds very closely to the test that we are conducting here, although the power is not particularly high (26%). These statistics are shown for each of the 43 counties in Table 2: the probability values of the Dickey-Fuller statistic were calculated from a Monte Carlo experiment with 60,000 observations. Only one Fisher test is significant out of 43, so we cannot reject the null hypothesis of a unit root.

17 Table 2. County Tests for a Unit Root. Bedfordshire 0.553 Essex 0.066 Monmouthshire 0.884 Surrey 0.388 Berkshire 0.386 Gloucestershire 0.984 Norfolk 0.427 Sussex 0.236 Buckingham 0.169 Hampshire 0.102 Northamptonshire 0.795 Warwickshire 0.835 Cambridge 0.792 Herefordshire 0.295 Northumberland 0.300 Westmorland 0.190 Cheshire 0.238 Hertfordshire 0.066 Nottinghamshire 0.559 Wiltshire 0.400 Cornwall 0.992 Huntingdonshire 0.594 Oxfordshire 0.675 Worcestershire 0.419 Cumberland 0.959 Kent 0.084 Rutlandshire 0.865 Yorkshire 0.693 Derbyshire 0.615 Lancashire 0.379 Shropshire 0.180 London 0.388 Devonshire 0.859 Leicestershire 0.713 Somersetshire 0.786 North Wales 0.236 Dorsetshire 0.907 Lincolnshire 0.694 Staffordshire 0.316 South Wales 0.835 Durham 0.042* Middlesex 0.552 Suffolk 0.611 Notes. Fisher tests for a unit root 1801-1820: for each county based on 20 (London 15 only) ADF tests of 30 observations each. Entries are p-values of the chi-squared statistic. Similar results are obtained for the period 1770 to 1820. For counties marked * the hypothesis of a unit root can be rejected at the five per cent level.

Proceeding on this basis, we imposed this restriction and estimated the further regressions

(5) ∆ ln Pt = α + λ1∆ ln Pt−1 + λ2∆ ln Pt−2 + ut

In the first case we can test for market efficiency by considering either the individual tests for λ1

= 0 and λ2 = 0 or the joint test λ1 = λ2 = 0: in all three cases we use robust standard errors with a small sample correction suggested by Davidson and McKinnon (1993) of multiplying the elements by T/(T – 2): the individual tests have a t-distribution and the joint test a chi-squared. In fact, tests for general heteroskedasticity using the White test, for ARCH and for autocorrelation suggested that the residuals were well-behaved and F and t tests for market efficiency using conventional moment estimators gave qualitatively the same results in all but a few cases. These provide strong evidence that we cannot reject the hypothesis of market efficiency once we have allowed for changes in the rate of return over time.

It would be possible to estimate the gross rate of return using α/(1 – λ1 – λ2) from equation (5). However, since we cannot reject the null hypothesis that λ1 = λ2 = 0, this method is unnecessary and will actually make the estimates less precise due to estimation error in λ1 or λ2. For this reason we in fact estimated the gross rates of return by just using the simple means of the weekly price rises. Tests for heteroskedasticity and autocorrelation suggested that there was no problem with this simpler specification.

18 4. The British banking system. The classic analysis of British banking in the eighteenth and nineteenth centuries is that offered by Bagehot (1873). Bagehot starts from a surprisingly modern standpoint, emphasizing the importance of capital in promoting economic growth: In countries where there is little money to lend, and where that little is lent tardily and reluctantly, enterprising traders are long kept back, because they cannot at once borrow the capital, without which skill and knowledge are useless. (p. 14).

Britain was in a fortunate position because: We have entirely lost the idea that any undertaking likely to pay, and seen to be likely, can perish for want of money; yet no idea was more familiar to our ancestors, or is more common now in most countries. A citizen of London in Queen Elizabeth’s time could not have imagined our state of mind. He would have thought that it was of no use inventing railways (if he could have understood what a railway meant), for you would not have been able to collect the capital with which to make them. (p. 7).

By contrast, Taking the world as a whole – either now or in the past – it is certain that in poor states there is no spare money for new and great undertakings, and that in most rich states the money is too scattered, and clings too close to the hands of the owners, to be often obtainable in large quantities for new purposes. (p. 7-8).

Bagehot argues that this ready availability of capital in Britain was due to her unique banking system: Concentration of money in banks, though not the sole cause, is the principal cause which has made the Money Market of England so exceedingly rich, so much beyond that of other countries. (p. 6).

Even the other advanced economies were then labouring under a rudimentary banking system and a shortage of capital: If you take a country in , even now, you will not find any such system of banking as ours. Cheque-books are unknown, and money kept on running account by bankers is rare. People store their money in a caisse at their houses. (p. 76).

So how exactly did the British banking system work? Joint stock banking was outlawed in England and Wales until 1826. The only exception to this was the Bank of England, which enjoyed privileged joint stock status in return for being the Government’s banker. Up to 1826, the banking system was therefore based on the country banks. These were private partnerships, restricted to a maximum of six partners who had to endure unlimited liability. Most banks were

19 consequently small and generally possessed only one outlet (i.e. there were very few banks with branches). To overcome their extremely limited geographic reach, the country banks developed correspondent relationships with a bank in London. In this way, they could send excess deposits to London and still earn a reasonable return on them; or they could borrow from London when they needed more funds locally. Hence country banks played the central role in recycling capital from surplus to deficit regions: [T]here are whole districts in England which cannot and do not employ their own money. No purely agricultural county does so. The savings of a county with good land but no manufactures and no trade much exceed what can be safely lent in the county. These savings are first lodged in the local banks, are by them sent to London, and are deposited with London bankers, or with the bill brokers. In either case the result is the same. The money thus sent up from the accumulating districts is employed in discounting the bills of the industrial districts. Deposits are made with the bankers and brokers in Lombard Street by the bankers of such counties as Somersetshire and Hampshire, and those bill brokers and bankers employ them in the discount of bills from Yorkshire and Lancashire. (p. 12).

Moreover, the country bankers were heavily engaged in this pursuit: All country bankers keep their reserve in London. They only retain in each country town the minimum of cash necessary to the transaction of the current business of that country town. Long experience has told them to a nicety how much this is, and they do not waste capital and lose profit by keeping more idle. They send the money to London, invest part of it in securities, and keep the rest with the London bankers and bill brokers. (p. 30-31).

In this way, the banks linked each of their localities directly to the London money market. It is interesting to note that there was little or no interaction between local banks. Instead there was a sophisticated network of bilateral relationships with London, and an almost complete absence of regional networks.

5. Gross Rates of Return and Financial Markets. In section three we estimated annual series for the gross rates of return on holding wheat in London and each county. In this section we consider the relationship between these rates of return and the presence of local banking services. To estimate this relationship as precisely as possible we need to consider how to measure the local supply of banking services, possible endogeneity problems between rates of return and the local supply of banking services, and possible alternative determinants of local rates of return.

20

5.1 The appropriate measure of banking services. Modern theories of banking tend to assume that firms operate within markets that are almost perfect and have a high level of information (Klein, 1976). In this context the usual measure of financial structure is the size of the money supply relative to GDP (which is just the inverse of the velocity of money). Alternative measures use variables such as credit or bank assets (Demirgüç-Kunt and Levine, 2001). However, apart from the fact that these variables are themselves endogenous, such data are not available for the period that we are studying. What is available in our period is the number of banking outlets operating in each county in each year. Shannon documented the number of banks operating in each county on an annual basis from 1801 onwards. We have consulted original documents to obtain further information on, not just the number of banking outlets, but the precise towns in which they were located, the number of banking outlets owned by each bank and the extent to which banks were related to one other (which is a different issue, as we discuss below). We propose to use the number of banking outlets as an indicator of the local availability of banking services. There are several reasons to suppose that this adequately reflects the local supply of banking services. First, notice that very few banks had more than one outlet. Although the ratio of banking outlets to banks was steadily growing over the period 1801 to 1815, it never exceeded 1.065. Dorset, the county with the highest ratio of outlets to banks, had only 15 outlets from 11 banks in the peak year of 1815. Over the entire fifteen-year period of our analysis, only five banks in the whole of England ever operated more than three outlets. So the number of local banking outlets was de facto the number of banks. In the econometric analysis below we report results only for the number of banking outlets; we repeated all of the analysis using the number of banks, rather than the number of banking outlets, and the results were virtually identical. Second, there is good reason to believe that the size of each banking outlet, in terms of the volume of banking services that it provided, was fairly constant across the sample. Country banks were typically very small. By law the maximum number of partners was six, but X partners was typical. The reason for this was that partners each had unlimited liability, so a banker would want only partners whom he could really trust and would lend only to borrowers about whom he had good information. This is reflected in the fact that the partners in a bank were often related to one another, either by

21 blood or marriage. The constraint on the number of partners constrained the size of the asset pool of each bank and ensured that the country banks were of a broadly similar size. The one potential countervailing factor came from the links between banks. Some individuals and some families were involved in more than one bank. In these instances, one might wonder whether two banks were really operating as independent banks, even though they were legally distinct and independent entities. We attempted to control for this by constructing a third measure of banks called “banking combinations” which takes account of identifiable linkages between banks. We repeated all of our analysis using banking combinations, rather than the number of banking outlets, and the results were virtually identical. Hence we do not report them here. Regardless of the whether we look at banking outlets, banks or banking combinations, we need to adjust the data to take account of the fact that our units of observation of interest rates vary greatly in size. The largest county (Yorkshire) is 40 times the size of the smallest county (Rutland). It is inevitable that Yorkshire will have more banks than Rutland, but would we expect it to have a smaller interest rate differential with London as a result? Obviously not, since the massive size of Yorkshire means that it would have taken a lot more banks to ensure that the local population had access to financial services. Moreover, the volume of business that each bank could undertake was limited owing to the legal constraints on bank size. So areas with a large and wealthy population were likely to demand more banking services and it is likely to take have required more banks to ensure that the local population had access to financial services. Perhaps an ideal measure would be to divide the number of banks (or outlets or combinations) by GDP. But such data are not available, so instead we divide by the local population total. These date are taken from the 1801, 1811 and 1821 censuses and interpolated using a cubic spline to avoid a kink in 1811.6 The simpler alternative was to divide the number of banks by the county area; the results were extremely similar to those reported here. Even after controlling for county size as described above, we were still not completely happy with an analysis based on counties. The rates of return that we estimated pertain to counties in the sense that they are based on changes in the official average price of wheat in each county. But in reality this average price is not based on every bushel of grain traded in a county.

6 The 1801 Censal data are adjusted for parishes known to be omitted following the suggestion of Wrigley and Schofield (1981). Linear interpolation of the data produces almost identical results.

22 Instead it is based on every bushel of grain traded in a subset of market towns in the county, where the minimum number of market towns in each county was three and the maximum was twelve.7 So what we are really observing is the average rate of return in the most important towns in each county (hereafter referred to as the ‘wheat towns’). Whilst we do not have the price data for each individual town (only the county averages were published in this period), we do at least know which towns were monitored in order to construct the average price. We use this information by breaking down the bank data for each county into banks that were situated in wheat towns and banks that were situated in non-wheat towns. We then re-estimated all of our equations using only wheat town banks as our measure of the number of banks in each county; in this way we hoped to reduce measurement error (and possible bias) arising from the fact that rates of return might be different (and possibly systematically different) in the wheat and non- wheat towns.

5.2 Alternative explanations for banking. In order to provide an adequate test of the effect of banking on the rate of return on grain, we need to consider two alternative potential explanations. First, transport costs were also falling over the period so that arbitrage in grain markets would have reduced the wheat price differentials across markets. Price differentials put a bound on the local interest rate. Take the extreme case where wheat markets were perfectly integrated; then there would be no scope for prices to differ and no differences in rates of return. Financial markets would effectively be integrated through the physical movement of commodities across counties. In fact, transport costs were sufficiently large that it would be possible for wheat prices to diverge in individual markets. Evidence from Jackman (1916, p.723) suggests that costs of carriage by road were about 0.75d per hundredweight per mile in 1810, corresponding to about 0.365d per bushel. Since the price of wheat ranged from about 50d to 150d per bushel, the cost of transporting wheat 50 miles (19d) was in the range of 13 to 38 per cent of the value of the wheat. However, it is certainly the case that there were improvements in transportation in this period. For example, the mileage of turnpikes increased from 15 833 miles in 1801 to 16 945

7 London had only one market that was officially monitored. For an exhaustive discussion of our data, see Brunt and Cannon (2001).

23 miles in 1815,8 and this might have been sufficient to narrow wheat price differentials enough to simultaneously erode rates of return on grain.9 Accordingly we consider the rôle of transport costs by looking at the density of turnpikes (miles of turnpike divided by area of county). When we consider town level data it is less obvious how to scale the mileage of turnpike, so we consider average mileage linked to a town and also a turnpike dummy.10

5.3 Endogeneity and measurement error. Both of our explanatory variables are liable to suffer from measurement error and endogeneity. The supply of banking services and the supply of roads is endogenous because we know that these were provided by the private sector in order to respond to arbitrage opportunities. Measurement error exists because our banks variable is only an imperfect measure of the supply of banking services and deflating by population does not fully take into account the potential demand for banking services. Similarly, the density of turnpikes does not fully capture the ease of transport. For both these reasons we should ideally use instrumental variable techniques. There are relatively few variables available on an annual basis for each county for the period 1801 to 1815. One variable that we use is the density of population in the locality (Census district) immediately surrounding the towns for which we have data. But this is available only for the censal years 1801, 1811 and 1821, interpolated by a cubic spline. The other data that we have available are information on newspaper circulation, both at county and town level. These are potentially good instruments because they would be correlated with the degree of development of each town while being unlikely to be correlated with either any measurement error or differences in rates of return. It is also extremely unlikely that such newspapers would have a direct effect on rates of return, since, so far as we know, they did not contain information on interest rates, nor information on prices of wheat except insofar as it was

8 These data were collected from Albert, The Turnpike Road System in England, 1663-1840; Pawson, Transport and Economy: the turnpike roads of eighteenth century Britain; British Parliamentary Papers 1821, vol. 4. 9 The other potential source of improvement in transport costs would have been improvements in the canal system, but virtually none of the towns from which grain prices were collected had a canal built in the relevant period 1801 to 1815, so there would be no within group variation. 10 We also considered dividing the mileage of turnpike by population, although this would only make sense if turnpikes were so heavily used that extra population required extra transport capacity. This alternative measure had the wrong sign with a coefficient that was both economically and statistically insignificant.

24 already available in the London Gazette.11 So these are good candidates for excluded instruments. They are also available annually for each county for the period with which we are concerned. Summary statistics of the main variables are presented in Table 3.

Table 3: Summary Statistics of the Main Variables. Mean Std. Dev. Overall Between Group Within Group Absolute difference in returns 0.233 0.188 0.058 0.179 County bank outlets per capita 0.063 0.030 0.025 0.017 County turnpike density 0.509 0.235 0.237 0.019 Town bank outlets per capita 0.237 0.123 0.104 0.066 Town turnpike miles 208.499 137.228 138.607 7.605

5.4 Results. All of the panel data analysis is based on fixed effects estimation since Hausman tests for the validity of random effects are overwhelmingly rejected. Most of the variation in the absolute differences in returns is within group variation, whereas most of the variation in the explanatory variables is between group estimation: this is particularly true for the measures of turnpikes. County-level regression results are reported in Table 4. In each case the bank variable is strongly and significantly negative: the effect of turnpikes is negative, although it is not statistically significant when IV or GMM is used. The OLS parameter estimates of about unity suggest that an increase of one standard deviation in our measures of banks and turnpikes would have led to a reduction in the differentials of rates of return of about 1.8 percentage points and 2.2 percentage points respectively, while the IV and GMM estimates suggest figures close to 3.8 percent points and 3.4 percentage points. Compared with an average absolute differential in rates of return in this period of 23.3 per cent, this is an economically significant if somewhat small reduction.

11 The exception would be those London newspapers that contained information on the price of wheat in London during the period when the London Gazette did not publish London prices. However, they contained no information on provincial wheat prices, only the London price. Interestingly, after 1815 even the London newspapers contained no information on London prices: all of the published data that we have found is simply for the national average price. Notice also that the London Gazette published the official prices one to two weeks in arrears, so that by the time they found their way into a local newspaper the data would have been well out of date.

25 Table 4: Regression Results from County-level Data OLS IV GMM County bank outlets per capita -1.040** -2.206* -2.195* (0.395) (1.047) (0.999) County density of turnpikes -1.182** -1.847 -1.762 (0.331) (1.295) (1.281) Centred R2 0.125 0.103 0.111 Uncentred R2 0.654 0.645 0.649 Number of Observations 615 615 615 Hausman test 42.93 34.67 χ2(2) [p = 0.00] [p = 0.00] Hansen J-test 2.407 0.511 χ2(1) [p = 0.30] [p = 0.47]

Notes: Dependent variable is the absolute difference in the return on wheat in London and each county (measured from January to April) from 1801 to 1815 for 41 counties, where North Wales and South Wales/Monmouth are each treated as a “county”. Estimation includes county fixed effects (not reported in the table). The Hausman test is the test for fixed effects being uncorrelated with the explanatory variables, i.e., that random effects estimation be valid. The Hansen test is the test for overidentification of the excluded instruments, i.e., that the instruments be valid. Instruments used are county area, county population and the number of newspapers in production/circulation within a county. Robust standard errors are in parantheses. Variables marked ** are significant at the one per cent level; variables marked * are significant at the five per cent level.

Over the fifteen years that we are analysing the number of banks increased by about 70 per cent compared with an increase of population of about 20 per cent. Since GDP growth per capita is unlikely to have been much more than 1 per cent per year, this means that the supply of banks relative to potential demand was about 30 per cent and it would be surprising if it had such a small impact on the differentials in the rates of return that Table 4 suggests. Table 3 shows that our the wheat towns had much greater access to banking services than non-wheat towns. The estimates in Table 5, based on the town-level data, suggest a much larger rôle for banks. The IV and GMM parameter estimates of about unity for banks suggest that an overall increase of banks by one standard deviation would have reduced differences in rates of return by more than 7 percentage points. The effect of transport costs in these regressions is much lower. The OLS and GMM estimates suggest that an increase in turnpike miles of one standard deviation would have reduced differences in rates of return by only 1.8 percentage points and 0.6 percentage points respectively. This may be due to the difficulty in reflecting the overall quality of transport at town level.

26 Table 5: Regression Results using Town-level Data OLS IV GMM Wheat town bank outlets per capita -0.239* -1.099** -1.210** (0.120) (0.320) (0.254) Wheat town turnpike mileage -0.0024* -0.0015 -0.0008 (0.0008) (0.0023) (0.0019) Centred R2 0.113 0.025 0.002 Uncentred R2 0.649 0.615 0.606 Number of observations 615 615 615 Hansen J-test 4.671 4.671 χ2(5) [p = 0.46] [p = 0.46]

Notes: Dependent variable is the absolute difference in the return on wheat in London and each county (measured from January to April) from 1801 to 1815 for 41 counties, where North Wales and SouthWales /Monmouth are each treated as a “county”. Estimation includes county fixed effects (not reported in the table). The Hausman test is the test for fixed effects being uncorrelated with the explanatory variables, i.e., that random effects estimation be valid. The Hansen test is the test for overidentification of the excluded instruments, i.e., that the instruments be valid. Instruments used are dividion density, town population, number of newspapers in production/circulation within the county, average and total number of newspapers in circulation within the towns. Robust standard errors are in parantheses. Variables marked ** are significant at the one per cent level; variables marked * are significant at the five per cent level.

What is consistent across Tables 4 and 5 is the evidence that banks had an economically and statistically significant effect in reducing the interest rate differential between each county and London. Turnpikes may have had smaller effect, although the evidence is much weaker. Moreover, when one considers the cost of setting up a bank versus the cost of building a mile of turnpike, setting up banks was clearly a more cost-effective way of achieving financial market integration. Obviously, building turnpikes had other benefits on the real economy, so constructing them was not necessarily irrational or unprofitable.

6. Conclusion and Discussion. In this paper we have analyzed county wheat prices in the period 1770-1820, concentrating on the period 1801-1815. The seasonal pattern of wheat prices is consistent with the model of intra-year storage posited by McCloskey and Nash. Hence we can reasonably attempt to estimate rates of return from that part of the year in which prices are rising. Our analysis also shows that markets were efficient over this period in the sense that prices followed a random walk with drift and the return on wheat was not serially correlated once changes in the rate of return had been taken into account. The rates of return on wheat in each county differ considerably from the rate of return on wheat in London. This suggests that regional capital markets were not well-integrated with

27 London because rates of returns were not fully arbitraged. Moreover, the differential between the county rates of return and the London rate is negatively correlated with the density of country banks. This is what we would expect as more entry into local financial markets led to more competition and a narrowing of margins for bankers. Consequently, the increase over time in the number of country banks reduced the interest rate differential between London and the counties, probably by around seven percentage points. Hence we can infer that banks were providing conduits for excess funds to be invested in London and enabling areas where credit was short to benefit from the London market. This is true even when controlling for alternative mechanisms that might have impacted on wheat prices, such as transport costs. This is strong evidence that county banks played an important rôle in facilitating financial market integration in the Industrial Revolution. Given the importance of finance in promoting economic growth, Britain’s exceptionally advanced banking system may well have been one of the key factors which distinguished her in the eighteenth and nineteenth centuries from competitor countries such as France and .

28 Bibliography. Albert W, The Turnpike Road System in England, 1663-1840 (Cambridge: the University Press, 1972). Allen, Robert C., ‘The Price of Freehold Land and the Interest Rate in the Seventeenth and Eighteenth Centuries,’ Economic History Review (1988), vol. 41, pp. 33-50. Arbuthnot, John, ‘An Inquiry into the Connection between the Present Price of Provisions and the Size of Farms,’ Annals of Agriculture (1796), vol. 27, pp. 8-41. Bagehot, Walter, Lombard Street: A Description of the Money Market (5th edition, London, 1873. Printed by Henry S. King). Barro, Robert J., “Government Spending, Interest Rates, Prices and Budget Deficits in the United Kingdom, 1701–1918,” Journal of Monetary Economics (1987), vol. 17, pp. 221 – 247. British Parliamentary Papers, “Evidence to the Select Committee on the Bank of England,” BPP (1819), vol. 3, p.291. British Parliamentary Papers, “Report on Turnpike Trusts,” BPP (1821), vol. 4, pp. 20- 340. British Parliamentary Papers, “Appendix No. 40 to the Report on Agriculture and Trade in Corn of the Continental States of Europe,” BPP (1828), vol. 18, pp. 284-9. British Parliamentary Papers, “Report on Turnpike Trusts,” BPP (1840), vol. 27, pp. 1-570. Brown, Robert L., and Stephen A. Easton, “Weak-form Efficiency in the Nineteenth Century: A Study of Daily Prices in the London Market for 3 per cent Consols, 1821 – 1860,” Economica (1989), vol. 56, pp. 61 – 70. Brunt, Liam and Edmund S. Cannon, “A Grain of Truth in Medieval Interest Rates: Re- examining the McCloskey-Nash Hypothesis,” University of Bristol Economics Department Discussion Paper No. 99-462 (1999). Brunt, Liam and Edmund S. Cannon, “Asset Prices in the Industrial Revolution: Evidence from the Grain Market,” University of Bristol, Department of Economics, mimeo. (2001). Buchinsky, Moshe, and Benjamin Polak, “The Emergence of a National Capital Market in England, 1710-1880,” Journal of Economic History (1993), vol. 53, pp. 1-24. Cameron, Rondo, Banking in the Early Stages of Industrialization (Oxford University Press, New York, 1967). Chambers, M. J., and R. E. Bailey, “The Price of Wheat in Early Modern England,” University of Essex Department of Economics Discussion Paper No. 438 (1995). Clark, Gregory, “The Cost of Capital and Medieval Agricultural Technique,” Explorations in Economic History (1988), vol. 25, pp. 265 – 94. Clark, Gregory, “A Precocious Infant? The Evolution of the English Grain Market, 1208– 1770,” mimeo. (undated), University of California at Davis. Crafts, Nick F. R., British Economic Growth during the Industrial Revolution (Oxford, 1986). Davidson, R., and J. MacKinnon, Estimation and Inference in Econometrics (New York: Oxford University Press, 1993) Dawes and Ward-Perkins Country banks of England and Wales : private provincial banks and bankers, 1688-1953(CIB Publishers, 2000) Demirgüç-Kunt, Asli and Levine, Ross (2001) Financial Structure and Economic Growth MIT Press Doornik, Jurgen A., OX: An Object-Oriented Matrix Programming Language (Kent: Timberlake Consultants Ltd, 1988).

29 http://www.nuff.ox.ac.uk/Users/Doornik/ Feinstein, Charles H., ‘Capital Formation in Great Britain,’ in Mathias, Peter, and Michael M. Postan (eds.) The Cambridge Economic History of Europe, vol. 7, part 1 (Cambridge, 1978). Fisher, R.A., Statistical Methods for Research Workers (Edinburgh, Oliver & Boyd, 1932). Ikhide, S. I., “Commercial Bank Offices and the Mobilisation of Private Savings in Selected Sub-Saharan African Countries,” Journal of Development Studies (1996), vol.33, pp. 117-132. James, John A., Money and Capital Markets in Postbellum America (Princeton, 1978). King, R. G. and D. Levine, “Finance and Growth: Schumpeter might be right,” Quarterly Journal of Economics (1993), vol. 108, pp. 717-737. MacKinnon, R. I., Money and Capital in Economic Development (Brookings Institute, 1973). Maddala, G. S., and S. Wu, “A Comparative Study of Unit Root Tests with Panel Data and a New Simple Test,” Oxford Bulletin of Economics and Statistics (1999), Special Issue,. McCloskey, D. N., and John Nash, “Corn at Interest: the Extent and Cost of Grain Storage in Medieval England,” American Economic Review (1984), vol. 74, no. 1 (March), pp. 174 – 187. Morton, John Chalmers, Hand-book of Farm Labour (London, 1868). Neal, Larry The Rise of Financial Capitalism (Cambridge, 1990) Pawson E, Transport and Economy: the Turnpike Roads of Eighteenth Century England (London, 1977). Pressnell, Leslie S. Country Banking in the Industrial Revolution (Oxford, 1956). Shannon, Herbert Austin, English Banking, 1800-1844, MSS., Nuffield College Library. Sial, Maqbool H. and Michael R. Carter, “Financial Market Efficiency in an Agrarian Economy: Microeconometric Analysis of the Pakistani Punjab,” Journal of Development Studies (1996) vol. 32, pp. 771-798. Taub, Bart, “A Model of Medieval Grain Prices: comment,” American Economic Review (1987), vol. 77, no. 5 (December), pp. 1048 – 1053. Townsend, Robert M., The Medieval Economy, (Princeton: Princeton University Press, 1993). Wright, John F., “British Government Borrowing in Wartime, 1750–1815,” Economic History Review, (1999) vol. 52, pp. 355 – 361. Wrigley & Schofield (1981) The Population History of England, 1541-1871 Young, Arthur, The Farmer’s Kalender (London, 1770).

30 Figure 3. The Change in County Bank Densities, 1800-20. Beds Berks Number of Banks per County Bucks Scaled by Population Cambs Cheshire Cornwall Cumberland 0.16 Derbs Devon Dorset Durham 0.14 Essex Gloucs Hants Hereford 0.12 Herts Hunts Kent Lancs 0.10 Leics Lincs Mids Norfolk 0.08 Northants Northumberland Notts Oxon 0.06 Ruts Salop Somerset Staffs Suffolk 0.04 Surrey Sussex Warks Westmorland 0.02 Wilts Worcs Yorks N Wales 0.00 S Wales/Mons 1800 1801 1802 1803 1804 1805 1806 1807 1808 1809 1810 1811 1812 1813 1814 1815 1816 1817 1818 1819 1820 31

24