It was seen to be obviously unsatisfactory to rely only on bottom-up engineering-based projections of efficiency gains because these would be likely to exclude both unanticipated gains and gains stemming from other sources, such as improvements in management or procurement. There has therefore been a more general debate on the efficiency gains a firm such as Railtrack might be expected to achieve, using what is described as a top-down approach.

European Economic Research Ltd. (EER) reported to the Rail Regulator on Railtrack's potential for efficiency improvement in December 1999. It considered Railtrack could reduce its expenditures (for constant output, i.e. excluding enhancements) by 3-5% per annum in real terms. Also in December the Regulator published provisional conclusions on revenue requirements in which he assumed savings of 5% a year, at the top end of the EER range.

Railtrack disputed this conclusion and produced a detailed response in April 2000 to which EER responded in May. Meanwhile Railtrack commissioned a study from OXERA to establish a cost-reduction target and the regulator commissioned NERA to look at railway experience in other countries, the latter in part in response to work on the United States by LEK that had been submitted by EWS. OXERA's work supported an efficiency target of 2%. NERA said that 3.3-3.9% can be regarded as an estimate of long term productivity growth in rail infrastructure and that Railtrack could also be expected to achieve additional improvement because of catching up to best practice.

I have read the EER, OXERA, and NERA reports as well as Railtrack's comments, and some of the academic literature referred to in these reports. This note does not attempt to comment on each of the reports in turn, but discusses the main arguments put forward. It therefore considers:

• The definition of efficiency improvement to be considered;
• The appropriate companies and sectors to be compared with Railtrack;
• The growth rates that emerge;
• The special factors claimed by Railtrack;
• The choice of the top end of the range; and
• The use of the assumption for the last two years of the present price control period.

As the various studies point out, the concept of efficiency to which the Regulator wishes to apply the assumption is Railtrack's total spending on its existing network. This includes both maintenance and renewal of the network. In considering the required level of maintenance and renewal expenditure, however, the Regulator intends to make separate allowance for the impact of growth.

This means firstly that the relevant productivity growth to be considered is akin to total factor productivity (TFP), rather than the return to one factor alone or to operating expenditure excluding depreciation.

When attempting to measure TFP it is necessary to consider the costs of the services of capital. This is not always easy because of the difficulty of valuing capital and of assessing the return it should earn and its rate of consumption (depreciation). Accounting rules are unlikely to accord precisely with an economic estimate of the cost. The factor inputs being considered under TFP should include both the cost of capital and its consumption. However, the concept to which the regulator is applying his efficiency assumption does not include the cost of capital and so the proper comparison is with an intermediate concept that includes capital consumption but not the cost of capital.

In Railtrack's case, capital consumption is assumed to be equal to the expenditure on renewing the system. This raises a familiar problem that expenditure may not be in steady state, and that is being addressed by other means. Moreover, most industries increase their capital intensity and this is almost certainly the case with Railtrack. Therefore renewal investment is likely to include not only capital consumption but also the investment required to increase capital intensity.

Thus the concept actually being considered is output relative to all factors less the financing cost of capital plus that part of "replacement" investment increasing capital intensity. The difference from TFP may be important, although the two differences may offset each other to some extent, and must be borne in mind when considering TFP results.

Secondly, since enhancements to the system are to be excluded and the costs of increases in use are to be recovered through other separate charges, the factor productivity assumption is to exclude the effects of economies of scale, and the studies make adjustments for this.

The studies derive assumptions for productivity growth that Railtrack can achieve from observations of what has been achieved elsewhere. In doing so they attempt to remove from the growth achieved elsewhere that element stemming from the exploitation of economies of scale. This is done by assuming an elasticity of costs with respect to output. If costs increase 0.9% for every 1% increase in output, a 10% increase in output will raise costs by only 9% and measured productivity will rise by almost 1%. Thus observed productivity growth must be adjusted by removing the 1% in order to estimate productivity growth with constant output.

There will be uncertainty over the correct elasticity to use in making the adjustment. Numbers as low as 0.2 have been cited for Railtrack, but these seem to be related to short run costs and not to growth over a sustained period. NERA's work on freight in the United States suggested elasticities of 0.87, although only about 0.3 if the extra volume was entirely absorbed in existing route miles.

In adjusting the figures for the industries to which it compares Railtrack OXERA assumes an elasticity of 0.9 throughout, although it argues that this is conservative. In a later paper, accompanying a Railtrack presentation on 22 June, they use 0.3 (but miscalculate the adjustment). In assessing the cost drivers in electricity distribution, a sector that is important in EER's comparison, the energy regulator has examined functions with an elasticity of operating costs (excluding depreciation etc.) of about 0.7 for most companies and 0.6 for those at the efficiency frontier.

A further observation should be made here. Given that the effects of economies of scale are to be removed from the efficiency assumption for Railtrack's existing system, they should be fully incorporated into the estimates for the cost of enhancements. When enhancements are costed individually the process of costing may well already incorporate the impact of economies of scale, but adjustments will clearly be needed when there are no individual cost estimates and perhaps even on some items when there are. Thus, for example, if an adjustment is made to assumed productivity growth using a cost elasticity of 0.8, the corollary is that enhancement costs per unit of output should be rated at only 80% of the level they would have represented if they had been part of the existing system.

Many cost estimates are made using unit costs but the units are normally of inputs and not of output. Economies of scale will be gained partly from economising on units of input (such as track) per unit of output and partly through reductions in the cost of acquiring additional units of input. Therefore input unit costs should not be scaled down by the full 20% (of this example) but by a smaller amount. The annual efficiency gain should be applied in addition to this adjustment.

The main difference between the conclusions of the various consultants stems from the differences in the main comparisons they have made. OXERA selects industry groups designed to match Railtrack's activities, EER selects recently privatised network industries, and NERA examines US railways. These are all comparisons of the rate of change of productivity. Cross-sectional comparisons of productivity levels are not used although NERA's work considered them.

This section discusses first NERA's cross-sectional work and then the various time series comparisons.

NERA concluded that the data they had obtained on UK, Canadian, Australian, Japanese, Swedish, and US railways could not support cross-sectional analysis. North American railways are predominantly freight and transport much larger loads per track mile than in the UK using fewer train miles per track mile. Japan has, for the most part, much greater passenger densities than the UK. The differences are so great that it is very difficult to tell whether any observed differences in efficiency from the UK are a result of the characteristics of the system or of better performance.

Where there is a degree of comparability, in Sweden and parts of Japan, NERA are not happy with the quality of the data they have obtained. This is unfortunate because the Hokkaido and Banverket systems seem at first sight comparable with Railtrack's, although much smaller. Hokkaido's way and structure costs per passenger mile are much lower than Railtrack's. Banverket's are higher, but it carries proportionately much more freight, which would tend to explain the cost difference.

However, the only detailed attempt to make an efficiency comparison between companies in different countries is LEK's comparison of freight costs in the US with freight charges in the UK. NERA reviewed that work and said "the general conclusion of the LEK analysis, that US Class I railroads have substantially lower costs than Railtrack, is broadly correct. It is far more difficult, however ... to draw any lessons ... about how much of this difference is due to inefficiency". It may not be the result of inefficiency but because Railtrack runs a larger number of lighter trains or because costs in the UK are raised by running a mixed passenger/freight network.

OXERA seeks to compare Railtrack with other sectors of the UK economy and derive a productivity growth rate that Railtrack might be expected to achieve. It analyses Railtrack's activities and assumes that what it does is roughly 40% like the utilities (which have experienced adjusted TFP growth of 21/2%), rather over 25% like services (¯ %), and a bit under 25% like construction (2%). The remaining 10% of Railtrack's activities are assumed to be like manufacturing or transport-and-communication (weighted average 2%). TFP growth is estimated over the period 1973-95, which does include some years of privatisation for the utilities but mainly results in a comparison that is not much different from one with the whole economy and so effectively treats Railtrack as like any other activity.

One might well argue with an analysis that assumes that Railtrack, a company that does not deal directly with customers, is significantly more like a financial services or retail company than are the utilities, which do. The assumption seems bizarre. It stems from a flawed analysis that separates Railtrack's activities into functions (such as signalling) and compares the single activity with an entire industry (such as gas, electricity and water). However, more important is the assumption that Railtrack is not in a special position as a recently privatised company and is not well placed to realise productivity gains from catching up to the efficiency frontier. This seems to be argued from two contentions.

The first contention is that, because Railtrack was created de novo "from a clean sheet, it was already more efficient than most other utilities."

There is no good evidence that utilities newly formed shortly before being privatised have been less able to achieve large productivity gains. Examples of companies in this position include the National Grid, National Power, PowerGen, and British Energy.

In Railtrack's case one could even argue for an a priori presumption of the opposite. The TOCs and leasing companies were carved out of the larger body for franchising or sale. It seems likely that residual elements would have been left behind in Railtrack rather than in British Rail, which would have been too small and non-operational to absorb them. Railtrack would then have even greater scope for cost reduction. Railtrack argues the opposite, that it had first choice of personnel and that the remainder fell to other companies, but, even in that case, it is not clear that Railtrack would have acquired the most adventurous staff. There can be no presumption that Railtrack is different from other privatised companies in this respect.

The second contention is that Railtrack was already as efficient as the private sector and/or that it has performed as well as privatised companies before being privatised. However, the latter claim is justified by claiming a comparatively low rate of TFP growth for privatised companies (and is therefore dealt with under that heading) and the former does not seem to be supported by analysis.

It seems to me that there remains a strong presumption that Railtrack, which has been privatised for a much shorter period than the comparators, has an opportunity for catch-up productivity growth like that achieved in other privatised companies. Indeed, it may even be in a better position to do so through being able to benefit from their experience.

EER compares Railtrack with gas and electricity transmission and distribution and with water and sewerage using recent data. OXERA cites (mainly) Parker and Martin, who compare other privatised companies in earlier periods.

The Parker & Martin study considers data up to 1995 and excludes the electricity and water data on which EER concentrates. The firms it considers include three regulated monopolies - BAA, BG, and BT - but are mainly firms in competitive industries that were privatised in the 1980s, then faced the 1988-92 recession, and had experienced a couple of years of recovery in the "latest" period.

The firms were:

• British Airways - privatised in 1987 after substantial changes produced by route liberalisation and the appointments of Lord King (1981) and Colin Marshall (1983);
• British Airports Authority - privatised in 1987 and whose performance since then cannot easily be assessed because the firm's output extends beyond air traffic movements; Parker and Martin use two very different output measures in their work;
• Britoil - a company in a special sector that was sold in 1982, adversely affected by the 1986 oil price fall, and absorbed by BP in 1988;
• British Gas - a company with no productivity growth after privatisation according to Parker and Martin but assessed by Waddams-Price and Weyman-Jones (Applied Economics 1996) as enjoying "6% annual post-privatisation productivity growth";
• British Steel - sold in 1988 after the MacGregor rationalisation of the early 1980s and severely affected by the 1990 recession;
• British Aerospace - created in 1977, sold in 1981, bought Rover in 1988, affected by the recession;
• Rolls Royce - sold in 1987;
• National Freight Corporation - sold in 1982, initially successful, but later squeezed by competition such as that from food companies' own fleets;
• Associated British Ports - privatised in 1983 but, unlike most of its competitors, still subject to the Dock Labour Scheme for many years;
• British Telecom.

Measurement of the factor inputs for these companies was not easy. Except in the case of British Gas only historic cost accounts were available. As a measure of the input of capital, therefore, Parker and Martin used recorded historic cost depreciation plus an 8% real "rental charge" on a measure (presumably also at historic cost) of the capital stock. This may have seriously affected the results, particularly at a time of decelerating deflation that would tend to result in rising historic cost capital/output ratios.

Even measurement of output caused problems, because it is not easy to distinguish changes in price from changes in quality (and so output). Parker and Martin quote different results for British Aerospace depending on whether turnover is deflated by the manufacturing producer price index or an index of input costs for the aerospace industry.

I consider the EER comparison to be with more obviously similar industries, which appear to have been less subject to the impact of special factors and to involve fewer measurement problems.

One problem with using post-privatisation data is a concern that costs were inflated at, or immediately after, vesting. If that were the case, EER's figures would show a decline from temporarily inflated figures. On the other hand, this might also be the case for Railtrack.

NERA has estimated equations explaining US freight productivity growth since 1986. They explain costs by a productivity trend, economies of scale (which increased 50% 1986-98), traffic density (which doubled), and firm dummies. In this way they attribute the rapid productivity growth in the period (when unit costs halved) between the three factors.

In discussing the results, I consider only the comparisons with recently privatised companies and US railroads.

EER claims that it observes a decline in operating costs excluding depreciation per unit of output of broadly 3-7% a year. However the range in its table 4.2, to which the claim refers is actually 3.7 - 9.1%. The 3-7% range effectively omits gas distribution, where output growth has been high, and rounds down at the bottom end. It then allows for (mainly) economies of scale and capital substitution and concludes that a 3-5% range is appropriate.

Electricity transmission and distribution, which are at the top end of the range with 6.8% and 6.5% p.a. improvement respectively if gas distribution is excluded, have experienced 1.4% p.a. output growth and water (as defined here) none. Capital substitution means that operating costs including depreciation have fallen less rapidly, by 1.6% for electricity distribution but perhaps 1/2-3/4% for water and sewerage. The range for this concept of productivity growth could then be assessed as 3.2-5.2%. If the elasticity of costs with respect to output were 0.8 the range adjusted for economies of scale would contract to 3.2-4.9%.

OXERA objects to the use of this restricted data set and relies heavily on Parker and Martin when considering recently privatised companies. As noted earlier the Parker & Martin data set includes British Gas in total but otherwise has no overlap with the EER set. EER quotes Parker & Martin as finding TFP growth for their "latest" period to have averaged 2.8% p.a. OXERA wishes to adjust this, by a) looking at Parker & Martin's "post-privatisation period" and b) excluding certain companies (Associated British Ports, British Steel, BAe, Rolls Royce, and BT) to derive a number near 2% or even below it.

As mentioned earlier, I do not regard the use of the Parker and Martin data as a good guide. They are not generally for firms with a marked resemblance to Railtrack and one could produce arguments for excluding any or all of them from the comparison, rather than just the companies selected by OXERA. The "latest" period is very short, three years or less for every company and often only two, and the "post-privatisation period", which consists in the four years following privatisation, often overlaps the "recession" period so that neither period used may be representative. There are serious measurement problems and the concept to which the figures are intended to apply is total factor productivity, including the return to capital. It would not be surprising if the measure that is relevant here (operating costs including depreciation) were to result in a significantly larger figure.

NERA's results for the United States show quite large scale and density effects, reducing the trend decrease in the relevant cost measure from a raw 6% or so to an adjusted figure of 3.3% per annum. It regards this as a steady state measure and considers that there is catch-up potential that can be added. It is interesting that the figure is not dissimilar to the O'Mahony steady state scale-adjusted UK TFP growth estimates of 2.6% for electricity, gas & water and 2.8% for transport & communication. It is slightly higher but, as the measure considered by NERA and by the Regulator excludes the return to capital (and capital substitution is likely to have raised that element), this is not surprising.

Comparison with recent utility experience gives a 3-5% range for adjusted productivity growth in the relevant concept and longer term analysis in the US and the UK suggests a steady state gain of over 3% to which a substantial catch-up element can be added. This appears to accord with O'Mahony's results and the Parker & Martin work does not seem to me to be inconsistent with it.

Railtrack suggests that a number of factors specific to the company reduce its potential for efficiency growth. These are that:

• The size of its projected investment programme will raise its costs;
• Improvements in output will be required;
• The scope for technical change may be less;
• Its labour intensity may make it liable to rising real input costs;
• The consequent scope for capital substitution makes a unit operating cost figure misleading; and
• The large weight of subcontracting in total expenditure makes the introduction of new practices and methods more difficult.

While there is a prospect of increased purchasing by Railtrack, it has been anticipated for some time, often involves skills not confined to the railway industry, and, where specialisation is called for, can resort to an international market. EWS have claimed that tender prices are falling at present.

Improvements in output are required from Railtrack but that was also the case for the industries with which comparisons were made in order to derive the efficiency increase assumptions. There should not therefore be an impact on the comparison. In addition, since the scale of the renewal programme for the next price control period is much greater than steady state, replacement of existing infrastructure with its modern equivalent should produce abnormal improvements in performance.

As EER points out, signalling, which forms a large part of Railtrack spending, seems to present a greater prospect of efficiency improvements as a result of technical change than is likely in the comparator industries. I understand that there is also significant change in other areas (such as high output track laying equipment).

The relative price of labour tends to rise as capital substitution occurs. However, there may also be greater scope for improvement in labour-intensive industries and I have seen no evidence of correlation between capital-intensity and TFP growth.

Indeed, labour productivity tends to rise faster than capital productivity and there may also be greater scope for capital substitution to increase overall efficiency.

One would have thought that subcontracting would make change easier. Rather than having to change practices within a single company, there is also the possibility of changing to a different company with better practices.

The regulator chose an efficiency assumption at the upper end of the range put forward by EER. This seems not unreasonable. The scope for catch-up might well be more than 1% a year (and so more than 6% in total by the end of the next review period) and, when that is added to a steady state of a little over 3%, the overall answer would be likely to be at the upper end of the EER range.

It is reasonable to apply the present best estimate for productivity growth potential from the present (i.e. including the last years of the present period) and not just from the start of the next period. That is because the objective is to forecast what an efficient company could achieve in the next period and, if such a company could make gains in advance of the next period, that is what should be assumed. There is no element of retrospection because revenue in the present control period will remain dependent on the assumptions made at the last review and will not be affected by any assumptions made now.

However, in assuming a starting efficiency and cost levels for the next review period, it will be important to ensure that volume changes between the last data point and the start of the review period are also costed at the marginal cost rates now deemed appropriate and not those assumed in the previous review.

I therefore conclude that:

• The appropriate concept to use is one akin to Total Factor Productivity adjusted to remove the effects of economies of scale but without the factor of the cost of capital;
• It should be assumed that Railtrack has an opportunity for catch-up productivity growth like that achieved in other privatised companies;
• 5% is a reasonable overall assumption;
• As part of the process of forecasting an efficient company's cost levels in the next price control period present efficiency assumptions should be used for the remaining years of this period and not the assumptions of five years ago; and
• Scale effects may need to be applied to the cost of enhancements.