I

In the aftermath of the panic of 1907, John Moody published his first railroadFootnote ² security analysis and ratings manual in April 1909.Footnote ³ The Poor Company followed in rating securities in 1916 and merged with Standard Statistics in 1941 to form Standard & Poor's (Sylla Reference Sylla, Levich, Majnoni and Reinhart2002). Sylla credits the advent of the securities rating industry as ‘a fusion of functions performed by’ the credit-reporting agency, the financial press and the investment banker. In this article we find support for this view, but also provide additional rationales for Moody's ratings approach, rooted in the economic and financial-market conditions that characterized the pre-Federal Reserve financial system.Footnote ⁴

Moody intended his first publication to be primarily a ‘complete explanation of the proper principles to be employed for analyzing railroad investment values’ (Moody Reference Moody1909, p. 14), and only secondarily as a ratings-based system of security analysis. Moody's intended audience was the vast number of smaller investors in railroad securities, since control of railroad firms was highly concentrated by 1909. As Moody states:

Moody focused his 1909 analysis on railroad securities, in part due to the illiquidity of the US Treasury market at the time, and thus in response to the demand for alternative high-quality securities. The US federal debt had declined sharply post-Civil War from a peak of about 40 percent of GDP in the late 1870s to less than 5 percent of GDP by 1910. As well, federal regulation required banknote issues to be 100 percent backed by eligible US government securities, which absorbed much of the outstanding US government securities (Friedman and Schwartz Reference Friedman and Schwartz1993, p. 23). The lack of a liquid government bond market increased demand for railroad bonds, which generally were considered high quality (i.e. near risk-free) by investors, as discussed further below.

Railroad bonds were common bank reserve investments at the time. For example, Hogan (Reference Hogan1913) cites 1912 Comptroller of the Currency data that, in the aggregate, national banks held $354,321,271 in railroad bonds, $179,322,004 in state, county, and municipal bonds, $195,452,530 in other public service corporation bonds, and $223,500,814 in all other bonds, while their aggregate capital stock was $1,046,012,500. As well, Nelson (Reference Nelson1907, p. 74) cites evidence from a 1905-6 survey of 7,000 banks and trust companies, which concluded that as much as 50 percent of individual bank reserves of eastern US banks were held in bonds, that eastern US banks preferred railroad bonds while western US banks preferred municipal bonds, and that all banks generally preferred listed bonds. Finally, A. G. Hoyt commented in Kemmerer's National Monetary Commission report (Reference Kemmerer1910, p. 217) that:

As well, railroad securities were preferred as margin collateral in the NYSE's call-loan market. Required margin on call loans was normally 25 percent of the loan amount, as long as (1) the securing collateral consisted of approximately two-thirds railroad and one-third industrial securities, and (2) these securities showed price stability and (more importantly) high liquidity (Griffiss Reference Griffiss1925, pp. 16-26). Sources of call-loan funds included the New York City banks, and the interior banks through their correspondent accounts with the New York City banks (Griffiss Reference Griffiss1925, pp. 11-14). These banks would pull funds from the New York call-loan market, and sell off secondary reserve holdings, as seasonal money-demand increased, causing call-loan rates to increase and pressuring bond and stock prices. For example, Seltzer and Horner (Reference Seltzer and Horner1922) showed an inverse correlation between the supply of call-loan market funds by banks and the call-loan rate. To be able to liquidate funds during times of crisis, such as the panic of 1907, the securities held as secondary reserves and as call-loan collateral needed to be of high security and salability.

The study addresses two interrelated issues with Moody's inaugural bond rating analysis. First, like other bond rating studies, the study investigates the extent to which Moody's ratings system reflected his personal information and judgment concerning the quality of railroad securities. For example, Pinches and Mingo (Reference Pinches and Mingo1973) remark that: ‘In addition to quantifiable data, the rater's qualitative judgment concerning the future ability of a firm to make interest and principal payments also influences the bond ratings.’ Pogue and Soldofsky (Reference Pogue and Soldofsky1969), in a study of corporate bond data from the 1960s, concluded that: ‘Although intangibles of judgment undoubtedly enter into the determination of bond ratings, the results of this study suggest that differences in bond ratings can be explained to a significant degree by available financial and operating statistics.’ The present study finds results comparable to Pogue and Soldofsky (Reference Pogue and Soldofsky1969), especially for the subsample of Moody's rated bonds which were followed in the financial press, but finds that Moody's analysis may have added information value for the roughly two-thirds of bonds in Moody's sample that were not reported on in the financial press.

However, security ratings can play important roles in addition to, or in place of, a private information role. Sylla (Reference Sylla, Levich, Majnoni and Reinhart2002) discusses the role that bond ratings might play in reducing agency conflicts, either between stakeholders and management, or between stakeholders, asset guarantors and asset managers. The Moody's quote above suggests a potential conflict between minority shareholders and the concentrated ownership of railroad firms, as will be further described in Section II below. Potential conflicts also existed between investors and other providers of railroad information and analysis, as with investment banker J. P. Morgan's controlling influence with many railroad lines. Finally, Smith and Walter (Reference Smith, Walter, Levich, Majnoni and Reinhart2002) hypothesize that rating agencies may provide informational economies by collecting and analyzing information for security investors more efficiently than investors could individually. The informational economics hypothesis is supported by the subscription model adopted by Moody, i.e. investors were the source of revenues to Moody's through their purchase of his manual.

Moody in 1909 rated 38.94 percent of railroad bonds in his sample as Aaa and 85.25 percent of railroad bonds as A, Aa or Aaa. Thus, the ratings helped investors identify which railroad bonds could be considered ‘near risk free’. To test whether the large percentage of high-rated bonds was justified, the study pursues a structural default-risk analysis to estimate the probability of default of railroad securities. The analysis will confirm that default probabilities were quite low, even during the panic months of 1907. Accordingly, Moody's manual makes clear that his 1909 security analysis and ratings system was intended to reflect an investment analysis, rather than a credit or default analysis, of railroad securities. More specifically, Moody's 1909 manual makes virtually no mention of railroad security defaults and associated credit risks.

Section II below discusses the cartelization of US railroads in the 1870-80s, the railroad industry consolidation that occurred with the economic depression of 1893, and the railroad regulation that started with the passage of the Interstate Commerce Act of 1887 and the creation of the Interstate Commerce Commission. This section sets the historical context to the advent of Moody's ratings. Section III then presents Moody's explicit system for constructing bond ratings based on: (1) the security of the issue, reflecting its security of principal and permanency of income, and (2) the salability of the issue, reflecting its liquidity.Footnote ⁵ While the security factor is a type of interest-coverage ratio, the salability factor was intended to reflect a security's liquidity, that is, the ability to sell the security without loss of value. It also discusses Moody's construction of the physical-factor, income-factor and capitalization-factor tables, which completes Moody's presentation of railroad system data for investors.

Section IV discusses the analytical basis for Moody's ratings definitions and the resulting distribution of Moody's railroad bond ratings. Moody's ratings definitions were based on his analysis of the risk factors that affected the pricing of railroad securities at the time. In particular, Moody's rating system was based on the dichotomy between (1) high-quality securities, i.e. securities whose prices were affected almost exclusively by interest-rate risk, and (2) lower-quality securities, i.e. securities whose prices in addition were affected by changes in the earnings power of the issuing railroad. It also presents the construction of common and preferred stock ratings.

Section V then describes the study's ordinal regression approach to analyzing Moody's bond rating assignments, and presents the resulting regression results. In the present case, the regression approach is simplified, since Moody's ratings construction was explicitly based on a bond's security and salability factors. The study incorporates these two explicit factors, plus other accounting and financial factors to test judgmental aspects of the ratings, including whether the panic of 1907 might have shaped the 1909 bond ratings. Section VI then presents a structural default analysis of the study's railroad firms. The default analysis is used to establish whether the high overall railroad ratings assigned by Moody can be justified. Section VII concludes.

II

Railroad cartels were organized in the US in the nineteenth century to fix rates among member railroad lines. One prominent cartel was the Joint Executive Committee (JEC),Footnote ⁶ which was organized in 1879 to formalize rate setting among railroad lines operating from Chicago and St Louis to the eastern US ports north of Baltimore (Ulen Reference Ulen1980). The JEC collected data on railroad traffic, provided arbitration to resolve disputes, allowed price adjustments in response to members’ cheating, and imposed sanctions on violating members. Porter (Reference Porter1983) analyzed data collected by the JEC over 1880-6, and characterized railroad behavior as switching between periods of collusive and non-cooperative behavior. The periodic breakdown in collusive behavior resulted in destructive price wars among railroad systems.

An economic depression in 1893 started another period of price wars and the failure of numerous railroads. The resulting reorganizations created consolidation within the industry (Kolko Reference Kolko1965, pp. 64-5). Moody (Reference Moody1904, pp. 431-3) describes how by 1904 some 95 percent of American railroad lines were controlled by the following six railroad groups:

1. (1) The Vanderbilt Group with $1,169 MM in capitalization and 21,888 miles of line,

2. (2) The Pennsylvania Railroad Group with $1,822 MM in capitalization and 19,300 miles of line,

3. (3) The Morgan Group with $2,265 MM in capitalization and 47,206 miles of lines,

4. (4) The Gould-Rockefeller Group with $1,369 MM in capitalization and 28,157 miles of line,

5. (5) The Harriman-Kuhn-Loeb Group with $1,321 MM in capitalization and 22,943 miles of line, and

6. (6) The Moore Group with $1,059 MM in capitalization and miles of line 25,092.

The financiers with dominant influence over these groups were: J. Pierpont Morgan, John D. & Wm. Rockefeller, W. K. & F. W. Vanderbilt, George J. Gould, A. J. Cassatt, James J. Hill, Edwin Hawley, H. H. Rogers, August Belmont, Thomas F. Ryan, and W. H. & J. H. Moore (Moody Reference Moody1904). As well, Moody (Reference Moody1904, p. 441) states that:

Railroad consolidation was intended to curtail the cheating behavior that had plagued the industry. Nonetheless, the rate wars continued. However, since the mid 1880s, the railroads had pursued federal help in controlling rate setting. Their political influence, and that of other railroad stakeholders, resulted in the passage of the Interstate Commerce Act (ICA) of 1887 and creation of the Interstate Commerce Commission (ICC).

The ICA was intended to regulate the monopolistic tendencies of the US railroad industry. Ulen (Reference Ulen1980) concluded that the ICA and the ICC was, at least initially, ineffectual in regulating the cartelization of railroad rates. In contrast, Prager (Reference Prager1989) finds that passage of the ICA supported the interests of the railroads, using event-study methodology which examined the price reactions of railroad stocks to news events related to the legislation.

While the ICC struggled with its authority in its early years, the Elkins Act of 1903 allowed the agency to impose fines on railroads offering rebates and required railroad companies not to deviate from published rates. Essentially, the Act accomplished what the railroad cartels could not in terms of fixing rates (Kolko Reference Kolko1965, p. 100). As Kolko (Reference Kolko1965, p. 117) states:

Therefore, railroad consolidation and federal regulation had combined to restore stability and profitability to the railroad industry, which then set the stage for Moody's analysis of the railroad industry. These two factors – consolidation and federal regulation – were most likely important in contributing to the low default risk of railroads at the time (see also the results of Section VI below), and thus the large number of highly rated railroad securities in Moody's rating universe.

III

Moody's 1909 railroad analysis focused on two primary characteristics of each railroad security: (1) the security of the issue and (2) the salability of the issue. The security factor is based on an analysis of a security's claim on railroad earnings relative to its required coupon interest (thus a type of interest-coverage ratio), and is intended to reflect the impact of railroad earnings fluctuations on security value.Footnote ⁷ The salability factor, on the other hand, was meant to reflect the security's liquidity, that is, the ability to sell the security without loss of value. These two characteristics – security and salability – were of particular importance to investors in 1909, as with the bank secondary reserve decisions and call-loan market conventions discussed above.

The construction of these two factors, and the related construction of bond ratings, is illustrated in Appendix I for the Cleveland, Cincinnati, Chicago & St Louis Railway. The first column in the table describes the particular issue. The second column (Lien on miles) describes the lien held by a particular issue, in terms of the priority of its claim and the miles of track held under the lien. Column 4 gives the interest required per mile of system (IR) for each particular security listed in the table. For example, in Appendix I securities (1)–(7) jointly required $581 (that is, $20 + $155 + $151 + $100 + $73 + $15 + $67) in interest payments against $2,894 in average income available (AIA, column 3), leaving $2,313 in average income available for more subordinate claims. The resulting factor of safety is given in column 5, which is calculated as (AIA-IR)/AIA (the interest coverage ratio) and equals 80 percent for the first seven securities listed in Appendix I.

The security and salability factors are then presented in Appendix I under the title Basis for rating. Bond ratings are given in the final column and are based on a security's security and salability factors. Of these two factors, the security factor appears to weigh more heavily in determining a particular security's rating, for example, as reflected in the ratings definitions given in Appendix III. As well, as emphasized by Moody:

An empirical analysis of these two explicit factors, plus factors introduced to reflect Moody's judgment, will be presented in Section V below. In the next section, the study will present the data series used in constructing the physical factors, income factors and capitalization factors. These series were useful to investors for analyzing trends in railroad performance, since Moody presented ten years of annual data for each variable. As well, the income factors data were used to calculate the average income available variable, which was then used to construct the security factor, as given above. These data series then complete Moody's presentation of railroad data, which made his manual comparable to other data-intensive railroad manuals of the time.

Moody organized his fundamental railroad analysis into three additional tables for each railroad system: (1) Table A. Physical factors presenting data on the railroad's physical assets and revenue sources, (2) Table B. Income factors presenting income-statement information, and (3) Table C. Capitalization factors presenting data on capital sources and their performance. Each table presents a 10-year data history, along with 10-year averages from similar properties of other railroad systems for comparison. Moody emphasized that the 10-year histories lent permanency and stability to his rating system, stating:

An example of the construction of these tables is given in Appendix II, again for the Cleveland, Cincinnati, Chicago & St Louis Railway.

The 10-year data histories presented in these tables were abstracted from Interstate Commerce Commission reports, and thus were taken from a quite public data source. As Moody states: ‘the records of the Interstate Commerce Commission embrace nearly all the facts that are necessary, but they are not presented in very satisfactory form for intelligent and accurate usage’ (Moody's 1909, p. 65). His comment reflects on the issue of Moody's intention in creating his security analysis and ratings system, and appears to lend support to Smith and Walter (Reference Smith, Walter, Levich, Majnoni and Reinhart2002)-type information economies that Moody could realize by providing an analysis of publicly available data.

The physical factors table (panel A of Appendix II) provides a 10-year history on track mileage, ownership of locomotives and passenger and freight cars, level of annual traffic and the resulting revenues. The data presentation would allow the investor to analyze trends in the railroad's physical expansion and trends in passenger and freight rates and revenues. In the case of the Cleveland, Cincinnati, Chicago & St Louis Railway, railroad trackage and equipment expanded along with train mile earnings, while its passenger and freight rates remained relatively flat.

The income factors (panel B) presents a 10-year history on gross and net earnings, maintenance and other expenses, common and preferred dividends, and balance carried forward. More importantly, the calculated total net income is carried over to Appendix I to calculate the factor of safety for the securities with the highest priority lien. As well, the margin of safety measure, defined as ‘the proportion of total net income remaining after payment of all current fixed obligations, including taxes, car trust principal and interest payments, miscellaneous items, etc.’ (Moody Reference Moody1909, p. 94), is a measure of the security of required interest and other charges.

The capitalization factor table (panel C) provides a 10-year history on major forms of financial capital outstanding, and several capital-performance measures, including margin-of-safety measures for preferred and common stock. The panel shows trends in the railroad company's capital structure decisions, along with trends in return-on-capital (net income on net capitalization).Footnote ⁸ Preferred and common dividends are listed next. Finally, the margin of safety measures are listed separately for preferred and common dividends, reflecting the extent that net income exceeded the level of preferred and common dividends paid, respectively. These dividend coverage ratios then apparently served as the basis for Moody's system of stock ratings, as further discussed below.

IV

Appendix III gives the ratings definitions used by Moody. Moody based the ratings definitions on an analysis of the risk factors that affected the pricing of railroad securities at the time. In particular, the rating system was based on a dichotomy between:

Accordingly, ‘A’-rated (that is, A, Aa and Aaa) security prices were mostly affected by interest-rate risk and little affected by the railroad's earnings-power risk, much like a ‘risk-free’ security, while ‘B’ rated securities were somewhat affected, and ‘C’ rated were largely affected (and thus ‘speculative’) by changes in a railroad's earnings power. Within a particular letter classification, the distinctions become more minor. Thus,

The distribution of ratings from Moody's 1909 rating system is given in panel A of Table 1. What is perhaps surprising is the large number of Aaa-rated securities: 38.94 percent of securities were rated as Aaa, the largest percentage of any rating category. In addition, 85.25 percent of securities were rated as A, Aa or Aaa. Panel A also shows the frequency distributions of the security factor and the salability factor. The overwhelming number of ‘A’-rated securities is in contrast to the more recent paucity of such securities, for example, as discussed by Vazza and Cantor (Reference Vazza and Cantor2002).

Table 1. Frequencies of ratings, security and salability

This distribution of ratings would imply that Moody generally regarded railroad securities as of high quality. Moody's view is partially supported by Hickman (Reference Hickman1958) (and as discussed in Sylla Reference Sylla, Levich, Majnoni and Reinhart2002), whose results showed that agency ratings (constructed as a composite average of the ratings of Moody's, Standard & Poor's and Fitch) performed quite well over the period 1900-43. The loss rates of Aaa, Aa, A and Baa (or their equivalents with other agencies), respectively, were -0.6, -0.4, -0.1 and -0.3 percent over the sample bonds’ life spans. Therefore, bond ratings appear to have performed well in forecasting subsequent security performance in terms of loss rates. The topic of the default risk in railroad securities at the time of Moody's 1909 analysis will be formally analyzed in Section VI below and further discussion contained in the study's conclusions.

In addition to the ratings distribution, panel B and panel C of Table 1 show the cross-tabulation of the bond ratings with the security factor and the salability factor, respectively. As would be expected, the concentration of ratings lies along the diagonal of the two tables, with the diagonal clustering more concentrated with the security factor, reflecting its greater importance in Moody's bond-rating outcomes.

Moody's 1909 ratings approach was inclusive of all railroad securities, i.e. railroad bond, preferred stock and common stock issues. Moody's rationale was simple:

Much less discussion of the establishment of common and preferred stock ratings is given in Moody's 1909 manual. Apparently, these ratings were based on the margin of safety calculations (a dividend coverage ratio) for common and preferred stock in the capitalization section, discussed above. Appendix IV gives the common stock and preferred stock ratings for each railroad company (when available), along with the range of bond ratings for the company. In general, a railroad's common stock rating is at the same level, or lower, than the preferred stock rating, which in turn is at the same level, or lower, than the lowest of the company's bond ratings, reflecting the pecking order of the claims of these securities. However, several notable exceptions exist in the table, perhaps for the reason given above by Moody that the modification of the terms of debt and equity contracts can alter the usual rank ordering of these securities.

V

Studies of bond ratings, such as Kaplan and Urwitz (Reference Kaplan and Urwitz1979), Pinches and Mingo (Reference Pinches and Mingo1973) and Pogue and Soldofsky (Reference Pogue and Soldofsky1969), use various financial and accounting factors, such as security subordination, interest coverage, profitability, firm leverage and issue or firm size, to replicate the ratings system used by Moody's and other rating agencies. However, as discussed above, Moody based his 1909 bond ratings approach on two explicit factors: (1) the security of the issue, reflecting its security of principal and permanency of income, and (2) the salability of the issue, reflecting its liquidity. Our analysis first investigates the extent to which these two explicit factors – security and salability – explain Moody's assignment of bond ratings.

In addition, Moody based ratings assignments on his personal judgment concerning individual securities. The study analyzes what factors might have shaped his judgment by incorporating two sets of covariates. First, the study investigates three accounting variables: (1) leverage: the ratio of long-term obligations (bonds outstanding plus capitalized leases) to total capital, (2) profit: the ratio of net income to total capital, and (3) size: total railroad system capital. Similar accounting variables have been predictive in other bond ratings studies. In addition, construction of these ratios is based on the 10-year average results presented in Moody's 1909 manual, to test whether Moody based his judgment on the long-run performance of these railroad lines.

Second, Moody's judgment also might have been shaped by how individual railroad securities reacted during the panic of 1907, since Moody's first ratings manual was published just some 18 months in the aftermath of the panic. The analysis examines whether bond ratings reflected the following characteristics of the 1907 panic year: (1) spread: a security's high minus low price during 1907 divided by the midpoint, a measure of the security's volatility, and (2) volume: the total volume of a security's trading activity during the panic months of October and November 1907. Data on (1) and (2) are taken from the January 1908 volume of the Commercial and Financial Chronicle (CFC). Note that these latter two variables are available only for bonds listed in the CFC.

An ordinal regression approach is adopted to account for the ordinal scale of the ratings. Let Y be the assigned bond rating (the ordinal response variable) with J = 11 potential categories: Aaa, Aa, A, Baa, Ba, B, Caa, Ca, C, D and E. The model assumes that values of Y are drawn independently from a multinomial distribution. Let γ_ij = Prob(Y ≤ j|x_i), j = 1 to J, be the cumulative probability of observation i being drawn from rating categories less than or equal to j, where x_i is a vector of independent variables. Note that γ_iJ = 1, hence only the first (J-1) γ's are needed in the model. Then γ_ij is given as:

• γ_ij = inv link[ (θ_j – β^Tx_i)/σ ]

• where

• inv link is the inverse function of the link function,

• θ is a vector of threshold parameters,

• β^Tx_i is the ‘regression’ part of the model, and

• is a scaling parameter.

In particular, a complementary log-log link function was used as the link function, and SPSS was used to perform the ordinal regression. In the ordinal regression analysis, the security and salability factors are categorical regression variables and the other financial variables are treated as continuous covariates.

The data analyzed by ordinal regression consists of the universe of railroad lines analyzed by Moody in the 1909 manual. A subsample of this universe was used to analyze the two panic variables: spread and volume, namely those securities listed in the January 1908 volume of the CFC, which published the high and low prices and NYSE trading volumes for railroad bonds over 1907. We refer to this subsample of securities as the: CFC-rated bond sample. The sample size of these two data samples, and the distribution of bond ratings for each, is given in Table 2. Table 3 then gives sample characteristics for the seven explanatory variables discussed above.

Table 2. Comparison of the Moody's bond universe and the CFC-rated bond sample

Note: Moody's bond universe includes all bonds rated by Moody as given in his 1909 manual. However, in contrast to Table 1, the CFC-rated bond sample is the sample of railroad bonds reported on in the CFC that were also rated by Moody in the 1909 manual.

Table 3. Sample characteristics of the regression covariates

Note: Moody's bond universe includes all bonds rated by Moody as given in his 1909 manual. The CFC-rated bond sample is the sample of railroad bonds reported on in the CFC that were also rated by Moody in the 1909 manual.

Net capital is taken from Moody's (1909) table C for each railroad line, and is defined as stock outstanding plus bonds outstanding plus rentals capitalized at 5 percent minus the company's Treasury securities. Leverage is calculated as (bonds outstanding plus rentals capitalized at 5 per cent) divided by net capital. Profit is taken from Moody's (1909) table C for each railroad line, where it is listed as net income on net capital. Volume is calculated as the sum of the volume of bonds traded, as reported in the CFC, for the weekly reporting dates: 4 October – 29 November 1907. Spread is calculated as the security's high price minus its low price for the year 1907 divided by the midpoint of this range. The data are taken from the CFC. As the results above illustrate, the ratings’ rank ordering roughly reflects the rank ordering of averages for size, leverage and profit variables, and inversely the rank ordering of averages for spread. Note also that the CFC subsample of bonds is about one-third of the size of Moody's rated bond universe, implying that roughly two-thirds of this universe is not reported on by the CFC.

Finally, to make the regression analysis more robust and more comparable across data samples, some ratings and salability categories with few observations were combined. Specifically, ratings below single B (that is, Caa, Ca, C, etc.) are combined into a single ‘sub-B’ category. As well, for the salability factor, categories M (medium) and P (poor) are combined with the category F (fair).

The left-hand-side (panel A) of Table 4 presents results of an ordinal regression of Moody's bond ratings on the security and salability factors, on which Moody explicitly based his ratings assignments. The data analyzed are the universe of railroad firms rated by Moody. As the results show, the overall fit of the model is significant with chi-square value of 1916.441 (with 8 degrees of freedom). The pseudo R-square measures range from 0.563 to 0.858, depending on the measure used, indicating that these two factors appear to explain a significant amount of the variation in the ratings assignment.

Table 4. Ordinal regression analysis of Moody's rated bond universe

The table presents an ordinal regression analysis of bond rating on (1) the security and salability factors, which were the explicit basis for rating assignments, in the left-side panel, and (2) these two factors plus three accounting covariates: (a) leverage, calculated as bonds outstanding plus capitalized leases to total capital, (b) profit, calculated as net income to total capital and (c) size, calculated as total capital, in the right-side panel. The data analyzed are the universe of bonds rated by Moody in his 1909 analysis of railroad securities. Finally, to make the regression results more robust, and more comparable across panel A and panel B, some ratings and salability categories with few observations were combined. Specifically, ratings below single B (that is, Caa, Ca, C, etc.) are combined into a single ‘sub-B’ category, which is the default category in the table above. As well, for the salability factor, categories M (medium) and P (poor) are combined with the category F (fair). Finally, since the ordinal regression models the cumulative probability, the ‘redundant parameter’ occurs at a cumulative probability of unity, i.e. at the highest rating of VH.

In panel B of Table 4, three covariates are added to the ordinal regression analysis of panel A, namely (1) size, calculated as total capital, (2) leverage, calculated as bonds outstanding plus capitalized leases to total capital, and (3) profit, calculated as net income to total capital. All three covariates are constructed from data on 10-year averages presented in Moody's 1909 manual. With these three additional covariates, the model fit is again statistically significant. The pseudo R-square measures range from 0.512 to 0.826, indicating approximately the same model explanatory power as with the previous regression model. The profit variable is significant at the 0.001 level and its negative coefficient indicates that more profitable firms achieved higher ratings (that is, ratings were coded in the analysis as 1 = Aaa, 2 = Aa, 3 = A, etc). The leverage variable is significant at the 0.01 level, and its negative coefficient indicates that more leveraged firms realized higher ratings. Finally, the size variable does not prove to be a statistically significant predictor of ratings assignment in this context.

The above two ordinal regressions are repeated in Table 5, but now using a subset of the study data that includes only those bonds that were rated and also listed in the Commercial and Financial Chronicle (CFC). This data subsample was selected to allow inclusion of two additional covariates related to the panic of 1907, namely (1) spread, which measures a security's volatility as the percentage spread of a security's high and low prices during the panic year of 1907, and (2) volume, which reflects a security's liquidity during the panic, measured as the total volume of a security's trading activity during the panic months of October and November of 1907. Data on these two variables are taken from the CFC.

Table 5. Ordinal regression analysis of the CFC rated bond sample

The table presents an ordinal regression analysis of bond rating on (1) the security and salability factors, which were the explicit basis for rating assignments, in the left-side panel, and (2) these two factors plus three accounting covariates: (a) Leverage, calculated as bonds outstanding plus capitalized leases to total capital, (b) Profit, calculated as net income to total capital and (c) Size, calculated as total capital, and two panic-of-1907 covariates: (1) PercentSpread, calculated as the percentage spread of a security's high and low prices during 1907, a measure of the security's volatility, and (2) TotalVolume, calculated as the total volume of a security's trading activity during the panic months of October and November of 1907, in the right-side panel. Data on these latter two variables were taken from the CFC. The data analyzed are those rated railroad bonds that were also listed in the CFC. Finally, to make the regression results more robust, and more comparable across panel A and panel B, some ratings and salability categories with few observations were combined. Specifically, ratings below single B (that is, Caa, Ca, C, etc.) are combined into a single ‘sub-B’ category, which is the default category in the table above. As well, for the salability factor, categories M (medium) and P (poor) are combined with the category F (fair). Finally, since the ordinal regression models the cumulative probability, the ‘redundant parameter’ occurs at a cumulative probability of unity, i.e. at the highest rating of VH.

In panel A of Table 5, the ordinal regression with just the security and salability variables is repeated with the CFC rated bond sample. Note in particular that the pseudo R-square measures now range from 0.642 to 0.891, indicating that these publicly followed and rated railroad bonds show somewhat more predictable behavior as reflected in the model's explanatory power. As shown in Table 2, the distribution of ratings is comparable in the two samples, so the ratings distribution is not a factor behind the greater explanatory power. Plausibly, the CFC sample is more transparent, in the sense that its published accounting and market information more accurately reflects the quality of these firms.

Finally, the regression analysis of panel A of Table 5 is repeated to include the three accounting covariates and the two panic-of-1907 covariates. The results are presented in panel B. The pseudo R-square measures now range from 0.941 to 0.999, again indicating that the publicly followed and rated railroad bonds show more predictable behavior. While the profit covariate is still highly significant: the size covariate is now significant in place of the leverage covariate. As well, the panic-of-1907 covariates are not significant predicators of rating assignment. One implication is that Moody based his ratings assignment on the two explicit factors: security and salability and the two ‘judgmental’ covariates: profit and size (in this case), while the performance of railroad bonds during the 1907 panic does not seem to have shaped Moody's judgment. All of these variables, except for the salability measure, are based on 10-year average results for each railroad system as presented in the manual. Thus Moody's rating assignments appear indeed to reflect the permanency and stability that Moody sought in his rating system.

VI

The large percentage of securities that were A-, Aa-, and Aaa-rated, as given in Table 1, suggests that Moody considered default probabilities of railroad securities to be too low to be the focus of his security analysis. Indeed, in Moody's 1909 manual virtually no attention was given to the issue of security default and the consequent losses. To further examine the issue of default probabilities, the study adopts an options-based procedure by Vassalou and Xing (Reference Vassalou and Xing2004) to extract default risk measures from the common equity prices of railroad firms, using daily equity quotes listed in the Commercial and Financial Chronicle. To construct a robust test of whether default risk was a factor behind security ratings, the study focuses on the year 1907, in particular to examine how the panic of 1907 impacted default measures, and conversely, the extent to which the panic of 1907 involved a solvency crisis among listed railroad companies.

Two default risk measures are presented by Vassalou and Xing (Reference Vassalou and Xing2004), namely, the ‘distance-to-default’ (DD), which measures the number of standard deviations from the current value of ln(V_A/X) to a value of zero (where default occurs), and the default likelihood indicator (DLI), both as given next.

• DD = [ln(V_A,t/X_t) + (μ-½σ²_A)T]/(σ_A√T)

• DLI = N(-DD)

• Where

• V_A,t is the value of assets,

• X_t is the strike price, that is, the level of liabilities due at the end of the estimation period,

• μ is the mean asset return,

• σ_A is the volatility of assets,

• T is the timeframe of the default likelihood estimation, here taken to be one year, and

• N(x) is the cumulative normal distribution to the point x.

Table 6 lists the sample of railroad firms used in this study and their Moody's bond ratings. The table also lists the minimum and maximum values of the distance-to-default (DD) measure over the study's sample period. Recall that DD measures the number of standard deviations of ln(V_A/X) to a value of zero As can be seen in the table, the default measure for most of the companies remained far from any solvency concerns throughout the crisis. The only railroad firm to reach within two standard deviations of default is Chicago & Alton Railroad, which in fact reached technical insolvency during the crisis. Therefore, these results appear to reject the hypothesis that railroad insolvency was a significant factor associated with the panic of 1907 and accordingly behind the construction of ratings by Moody in 1909.

Table 6. Distance-to-default (DD) measure results

The table presents results from a Vassalou and Xing (Reference Vassalou and Xing2004) structural default-risk analysis to extract default risk measures from the common equity prices of railroad firms with daily equity quotes. Equity data on railroad firms are taken from CFC. The DD measure gives the estimated distance-to-default, in terms of the number of standard deviations (st. devs.) from the current value of ln(V_A/X) to a value of zero (where default occurs). DD is measured as [ln(V_A,t/X_t) + (μ-½σ²_A)T]/(σ_A√T), where V_A,t is the value of assets, X_t is the strike price, that is, the level of liabilities due at the end of the estimation period, μ is the mean asset return, σ_A is the volatility of assets, and T is the timeframe of the default likelihood estimation, taken to be one year.

Min is the minimum, and Max is the maximum, distance-to-default (DD) measure for each stock over the period: 29 June 1907 – 27 Dec. 1907.

VII

Moody's first publication of a security analysis and ratings system in 1909 spawned a highly successful ratings industry. Sylla (Reference Sylla, Levich, Majnoni and Reinhart2002) credits the advent of this industry as a fusion of functions performed by the financial press, the investment banker and the credit-reporting agency. Moody's approach to security ratings featured: (1) an explicit security analysis system based on analyzing the security and salability of railroad securities, (2) a rating system which served to signal Moody's security appraisal to investors, and (3) a focus on 10-year data histories that allowed investors to analyze trends in the railroad industry, much as with other financial publications.

Moody may have intended his ratings to serve as a mechanism for conveying his private information to investors. As well, without relying on a private-information explanation, the ratings might have been intended to serve in an agency role, where ‘the ratings solved a principal-agent problem between investors and company managers’, or alternatively, that ratings ‘help to resolve conflicts of interest that might otherwise exist among the owners of financial assets’ (Sylla Reference Sylla, Levich, Majnoni and Reinhart2002, pp. 26-7). For example, Moody's ratings might have served to protect minority shareholders from the concentrated ownership of US railroads at the time. Finally, Smith and Walter (Reference Smith, Walter, Levich, Majnoni and Reinhart2002) hypothesize that rating agencies may provide informational economies by collecting and analyzing information for security investors more efficiently than investors could individually.

To test these hypotheses the study adopts an ordinal regression analysis of Moody's bond ratings. The basic regression analysis incorporates Moody's two explicit factors, security and salability. Moody based his ratings construction on these two explicit factors plus his judgment concerning the securities. In the resulting regression analysis with these two explicit factors, the resulting pseudo R-square measure ranged from 56.3 to 85.8 percent depending on the specific measure used. Therefore, these explicit factors account for a large portion of the ratings variability.

To test the basis of Moody's judgment, additional implicit factors were incorporated into the ordinal regression. The covariates: size (calculated as total railroad capital), leverage (calculated as bonds outstanding plus capitalized leases to total capital) and profit (calculated as net income to total capital) were constructed from 10-year average data presented in Moody's manual. The regression results indicate that the profit and leverage covariates were significant, while the size covariate was not a significant predictor of ratings, when combined with the two explicit factors. However, the regression pseudo R-square does not improve over the regression model with the two explicit covariates. Therefore, these additional accounting ratios do not add explanatory power.

To further test the basis of Moody's judgment, two additional variables are constructed that reflect financial-market conditions during the panic year of 1907. This panic occurred some 18 months prior to the publication of Moody's first manual, and security performance during the panic may have shaped Moody's judgment. These panic-of-1907 variables are (1) spread: a security's high minus low price during 1907 divided by the midpoint, which reflects the security's volatility during 1907, and (2) volume: the total volume of a security's trading activity during the panic months of October and November 1907, which reflects the security's liquidity during the panic months. These two covariates did not prove to be significant, suggesting that Moody's judgment was not shaped by the panic conditions of 1907. Moody's stated intention was to construct a system of ratings that reflected ‘the element of permanency and stability’ (Moody Reference Moody1909, p. 61) and the lack of significance of the panic-of-1907 covariates gives support to the statement.

The regression results also show that bond-rating predictability is much greater for the subsample of rated bonds (365 out of 1,137 bonds or 32.1 percent of the total rated-bond sample) that also were quoted in the Commercial and Financial Chronicle. The pseudo R-square measures for this sample of 365 railroad bonds range from 94.1 to 99.9 percent, while for the universe of rated bonds the pseudo R-square measures range from 51.2 to 82.6 percent. The implication would seem to be that railroad bonds followed by the financial press were much more transparent in terms of the accounting and liquidity measures of the issuing company. Therefore, the advent of Moody's bond ratings most likely had less significance for securities followed in the financial press and greater impact on the much larger sample of securities without financial press coverage.

Furthermore, Moody based much of his explicit analysis on 10-year data histories that were abstracted from Interstate Commerce Commission reports, a very public data source. As Moody states: ‘the records of the Interstate Commerce Commission embrace nearly all the facts that are necessary, but they are not presented in very satisfactory form for intelligent and accurate usage’ (Moody's 1909, p. 65). His comment would seem to indicate that Moody felt the contribution of his security analysis and rating system was: (1) the Smith and Walter (Reference Smith, Walter, Levich, Majnoni and Reinhart2002)-type information economies that accrued from aggregating and organizing the presentation of public data, and (2) his security-analysis approach which focused on the security and salability of each security issue as an explicit basis for assigning ratings. As discussed above, the security and salability factors do indeed account for a substantial amount of the variation in ratings.

Moody in 1909 rated 38.94 percent of railroad bonds in his sample as Aaa and 85.25 percent of railroad bonds as A, Aa or Aaa, underscoring his opinion that railroad bonds were generally high-quality securities. In choosing the railroad industry, Moody was able to identify a class of nearly ‘risk-free’ securities that could serve as substitutes for the illiquid Treasury market of the time. However, Hickman (Reference Hickman1958) (and as discussed in Sylla Reference Sylla, Levich, Majnoni and Reinhart2002) presented results that indicate that the overall default rate of the railroad industry was higher than with other industries over the period until the time of the 1958 study. The higher default rate of railroad bonds may have reflected the many intervening events since Moody's 1909 analysis, such as a period during World War I when the railroads were nationalized, increased competition between trucking and railroads, the Great Depression, and regulatory changes that may have disadvantaged the railroads relative to other transportation modes.Footnote ¹⁰

To better understand the issue of default risk and how it related to Moody's ratings construction, the present study employs a structural default analysis by Vassalou and Xing (Reference Vassalou and Xing2004). The analysis indicates that the majority of railroad securities faced negligible default risk, even during the panic months of 1907. These results lend support to the discussion above that the railroad industry had achieved a great deal of stability and profitability by 1909, through consolidation within the industry and through federal regulatory efforts to stabilize railroad rates.

An implication of the structural-default analysis results is that Moody was perhaps correct not to focus his analysis on the default or credit risk of the railroad bonds he rated. In fact, Moody makes virtually no mention of the default issue in his manual. As an alternative explanation, the results of Elton et al. (Reference Elton, Gruber, Agrawal and Mann2001) show that much of a bond's yield spread is indeed unrelated to default risk, and is rather related to the same risk factors that affect equity pricing. Moody's security-analysis approach seems to capture the intuition of this result, since Moody focused on an investment (rather than a credit or default) analysis of railroad securities, which, as well, was conceived as a common framework for analyzing railroad bond, preferred stock and common stock issues. Moody's rationale was simple: