The diversity among trust companies in New York City is a recurrent issue in this narrative. It underpins the difficulty with which J. P. Morgan organized collective action among the trust companies to assist the most‐distressed firms. The nature of this diversity informs the thesis that financial instability broke out in the periphery of the financial community—that is, by their absence from the New York Clearing House, the trust companies set themselves on the periphery of nationally and state‐chartered banks. The narrative also suggests that there was even a periphery within the periphery (the Heinze–Morse affiliates), which was where the trouble really began. Finally, there was an aspect of this diversity among the trust companies that would later appear in the Money Trust hearings, namely, that the financial elites in New York (the Morgan–Baker–Stillman circle) exploited the Panic for their own benefit. Recent studies on trust company performance during the Panic raise insights that highlight sources of diversity and invite future research.

Sources of Division: Business Model, Market Power, Exposure to Market Discipline

What can the movement of deposits during the Panic tell us about the relative significance of sources of division within the Manhattan trust company community? This appendix integrates and examines three perspectives as a supplement to the discussion of trust company diversity in previous chapters.

1. Business model. Hansen (2014) distinguishes recent entrants from long‐term incumbents in the trust company field based on business model and location: “uptown” versus “downtown.” He observed that the uptown trust companies tended to be smaller, younger, and more retail‐oriented, with smaller average account sizes and larger numbers of deposit accounts. In contrast, downtown trust companies tended to have a wholesale orientation, treating commercial clients with larger accounts who may have been less prone to run in a financial crisis. The distinction reflected the clientele they served (commercial firms versus individuals), the value proposition they offered (e.g., convenient location for individuals living uptown), and service mix (e.g., investment management for families and a greater emphasis on checking accounts for individuals, even checking accounts for children). Despite smaller average deposit balances, the uptown firms could prosper nonetheless if they attracted more customers, which they were doing at the expense of commercial banks.
2. Market power. The Pujo Committee alleged the existence of an oligopoly—a “Money Trust”—that exploited market power at the expense of customers and conspired to restrain competition. Senator La Follette and others directly charged that the financial elite used the Panic to discipline and suppress the aggressive growth of the trust companies in New York City. Given less attention in 1907 was the possibility that trust companies affiliated with the financial elite might have benefited from the Panic at the expense of unaffiliated trust companies. A study by Fohlin and Lu (2021) asked whether “opportunistic business adversaries took advantage of conditions that were ripe for a rumor‐fed panic … those connections may have provided some business advantages and potentially a liquidity backstop to those connected trusts.”¹ By studying the change in stock prices of trust companies (not the change in deposits) during the Panic, the authors found that “the connected trusts indeed benefited from their connections … the connected trusts rebounded much faster after the intervention by Morgan and the Treasury.”²
3. Exposure to market discipline. Hansen (2014) pointed out that losses of deposits were concentrated at four trust companies: the Knickerbocker, Trust Company of America, Lincoln, and Fifth Avenue. Frydman, Hilt, and Zhou (2015) wrote, “The deposit losses … were strongly influenced by any observable connection to Charles Morse or the other speculators involved in the failed corner.”³ They found that the deposits of trust companies having an affiliation declined by 53 percent, as opposed to only 23 percent for those with no affiliation.⁴ This is consistent with reports in contemporary newspapers and implies that the runs on those institutions were instances of depositors monitoring and disciplining bad actors by means of exercising the right to withdraw deposits. Therefore, perhaps the pattern of deposit losses had less to do with business model or market power, and more to do with the exercise by depositors of monitoring and market discipline. The thesis of market discipline has a long pedigree. For instance, Gorton and Mullineaux (1987) wrote,

A “bad” bank’s failure or suspension, for example, would induce bank customers to monitor the quality of their own bank’s liabilities. The cheapest way to monitor was to exercise the deposit contract [i.e., to withdraw deposits]… . A banking panic may be seen as an instance of customer monitoring. Exercising the deposit contract’s option feature en masse represents a cheap way for bank customers to monitor the ability of their bank to perform, and in effect, to monitor the monitoring of the [clearing house].⁵

These three explanations are not mutually exclusive. Nevertheless, integrating them in an examination of deposit changes in 1907 might help to illuminate the relative significance of any of the three explanations as a stimulus to future research. We offer the following analysis as a preliminary illustration of sources of diversity and of the challenges of interpreting those sources.

Data and Variables

We hand‐collected data on trust company attributes from reports of the New York Superintendent of Banks⁶ and focused on the 38 Manhattan‐based trust companies described there.

Dependent variables. For robustness, we studied six dependent variables: the value and percentage change in the number of deposit accounts across 1907, the value and percentage change in the dollar value of deposits across 1907, and the value and percentage change in the dollar value of deposits from August 22 to December 19, 1907.

With respect to independent variables, three general factors motivated their selection and inclusion:

1. Business model descriptors included measures of each trust company as of January 1, 1907: the number of deposit accounts, average dollar size of the account, age of the firm, and location. Hansen’s discussion suggests that retail‐oriented trust companies were more vulnerable to depositor runs. Such firms tended to be younger, have a larger number of deposit accounts, smaller average deposit balances, and were located closer to their retail clientele, uptown. The age of the firm was estimated as the number of years since chartered by New York State. Regarding location, we used three variables. Two were dummy variables to indicate the location of the trust company headquarters as uptown or downtown—for a robustness check, one dummy variable assumed the dividing line was 14th Street, and the other assumed Chambers Street.^a As a third measure of location, we developed a continuous variable based on distance from Wall Street, estimated^b as the shortest number of miles to walk to the headquarters of the trust company from the intersection of Broad and Wall Streets (the traditional heart of the financial district and the location of the New York Fed, the NYSE, and the headquarters of J.P. Morgan). If business model matters in explaining differences among the changes in deposits of trust companies, and if retail‐oriented trust companies were more vulnerable to runs than wholesale firms, then variables associated with a retail model should be significant.
2. Market power measures included market share of deposits among the Manhattan trust companies. In the industrial organization literature, high‐share firms are deemed to have more power to set prices and quantities (Bain, 1956). And consistent with the Money Trust theories, we included two dummy variables equaling 1 if the board contained familiar representatives or affiliates of J. P. Morgan, James Stillman, or George F. Baker, and zero if not, and the second to reflect whether the trust company maintained an affiliation with the NYCH (value of 1 if affiliated, and zero if not). Did a large market share and such affiliations serve to quell the loss of deposits? If so, the effects of market power should be significantly positive.
3. Monitoring and discipline. Did affiliation with Heinze and Morse overshadow other possible explanations for the change in deposits of trust companies? The list of affiliated trust companies given in note 28 in Chapter 7 was the basis for a dummy variable that took the value of 1 if the trust company were affiliated and zero if not. If the deposit changes are substantially a matter of monitoring and discipline, the association between the dummy variable and outcome measures should be negative and significant. A subset of the trust companies with an affiliation to Heinze and Morse experienced severe runs (Knickerbocker, Trust Company of America, and Lincoln).

Descriptive Statistics

Table A.1 gives a summary of the variables in this study. Several insights stand out. First, the Panic took a toll on the dollar value of deposits across the sample (lines 3–6). For the year 1907, the volume of deposits for the Manhattan trust companies declined on average nearly 21 percent (line 4); for the intense episode, deposits fell on average nearly 32 percent (line 6). However, the change in the number of deposits is more nuanced. We do not have data for the August–December period; but for the entire year, the percentage change in the number of deposit accounts was positive 6.8 percent on average, reflecting outsized percentage gains at a few of the smaller trust companies (line 2, and visible in Figure 16.2). But the absolute change in the number of deposit accounts was negative during the year (line 1). The minimum, maximum, and standard deviations of the deposit variables (lines 1–6) suggest great variation among the trust companies. This underscores the diversity among trust companies that frustrated J. P. Morgan.

Confirming the asymmetric size distribution of trust companies in Figure 16.1, lines 7 and 8 of Table A.1 reveal substantial variation in the number of deposit accounts and average dollar value of deposits. The measures of location (lines 9–11) show that about one‐quarter of the trust companies was located uptown, consistent with the map in Figure 9.1. And the average age of the sample of trust companies, 21 years (line 12) underscores the notion that these financial institutions were somewhat more recent additions to the financial community than were national banks, the earliest of which were founded in 1863. The statistics about market share of deposits (line 13) suggest that the trust company industry was not very concentrated, with an average share of 2.6 percent and a maximum share of 10 percent.

Correlations

Table A.2 gives the bivariate correlations among variables in the study, transformed where appropriate.^c Of greatest interest are the correlations of independent variables with dependent variables—the lower left quadrant (lines 7–17, columns A–F). In this quadrant, four independent variables stand out for their large and significant correlations with deposit changes.

First, the size and significance of the number of deposit accounts at the start of 1907 (line 7) is arresting: all the coefficients are negative and two are significant. Firms with a large number of deposit accounts suffered larger runs. Such firms would be the retail‐oriented trust companies, suggesting that business model was indeed an important factor in the runs.

Second, affiliation with Heinze and Morse (line 14) is also uniformly negative, for which four of the six are significant. This is consistent with the monitoring and discipline hypothesis.

Third, market share (line 13) has two large and significantly negative coefficients. This contradicts the notion that market power (share of market) was strategically useful during the panic. Knickerbocker ($56 million in deposits) and Trust Company of America ($46 million) were, respectively, the second‐ and fifth‐largest trust companies by dollar value of deposits at the start of 1907. The deposit losses of those two firms materially influenced the two large and significantly negative correlation coefficients. Thus, the results in line 13 may offer more support for the market discipline hypothesis than for the market power hypothesis.

Fourth, the “dog that did not bark” was the influence of financial elites, gauged either by board representation or by the trust’s affiliation with the NYCH (lines 15 and 16). The absence of significant coefficient coefficients with the dependent variables suggests less influence for the Money Trust hypothesis.

Finally, the lower right‐hand quadrant (lines 7–17 and Columns G–Q) reveals collinearity among a number of the independent variables. Some of this is intuitively obvious: the measures of location would likely be related; the three most severe runs (Knickerbocker, Trust Company of America, and Lincoln) were located uptown, linking the variable of severe runs with uptown location; older trust companies tended to have larger average account sizes, either by virtue of their years in service or their focus on the wholesale model. These correlations underscore the challenge in attempting to isolate the sources of division among the trust companies in Manhattan. Multicollinearity renders statistical inferences about independent variables less reliable—it does not negate the ability to make any inferences. But it does caution the analyst to look actively for contingencies and interdependencies, to use econometric remedies where appropriate, and to vet quantitative estimates against other information where possible.

With caution, we find that the correlation coefficients are consistent with the notions that differences in business models were the dominant source of division and that affiliation with bad actors would expose a trust company to serious discipline. On the other hand, the correlation coefficients are not consistent with the thesis that market power (share of market, affiliation with the financial elite, or membership in the NYCH) would yield solid benefits in deposit changes during the Panic.

Regression Analysis

One approach to working with collinear data is to employ log‐log stepwise regression analysis, suited to select independent variables both for their explanatory power and minimization of collinear effects.^d Table A.3 summarizes the results of stepwise regressions of the 11 independent variables against each of the six dependent variables. The collinearity statistics reveal low levels in the resulting models. The large and significant F‐statistics show that five of the six models fit the data better than an intercept‐only model (in Panel 4 the stepwise algorithm did not include any variables).

Of interest are five variables that stepwise regression selected. First is the dummy variable indicating an affiliation of the trust company with the Heinze–Morse circle (Panels 1, 3, 5, and 6). In all of these cases, the coefficients are negative and significant. This dummy variable is the most prevalent of the variables selected in the regressions.

The next two variables, the number of deposit accounts at the start of the year (Panels 1 and 2) and average value per deposit account at the start of the year (Panel 6), are indicators of the business model that the firm pursued. Large numbers of deposit accounts with relatively small account balances are consistent with the business model of retail‐oriented trust companies. Consistent with this, the coefficient for number of deposit accounts is significantly negative—a greater number of deposit accounts is consistent with the retail business model, which suffers greater deposit loss during the Panic. And the coefficient for size of deposit balances is significantly positive—larger value per deposit account is consistent with a wholesale model, which suffers less deposit loss during the Panic.

The fourth variable of interest, distance from 23 Wall Street, appears in two estimates (Panels 3 and 5). The coefficients are positive and significant. After controlling for affiliation with Heinze and Morse, greater distance from the financial district is associated with increases in deposits. This result is consistent with Figure 16.2, which showed that three trust companies experienced large increases in the number of deposit accounts in 1907—together they averaged 2.9 miles to the north; and all three were small and young firms, averaging 1.6 years. Thus, the positive coefficients on distance from 23 Wall Street sustains Hansen’s business model hypothesis in an unexpected way: the difference in business model from the downtown firms—perhaps proximity to their customers—seemed to yield an advantage in deposit changes during the panic.

The fifth variable, the dummy indicating affiliation with the NYCH, appears only in Panel 6 and shows a positive and significant coefficient. In this instance, it appeared that such affiliation yielded benefits. This is a surprising result, given the absence of significant correlation coefficients for this variable in Table A.2 (line 16).

Discussion

In general, these results build on previous studies and sustain at least four insights (with appropriate caution). First, differences in business models (wholesale versus retail) mattered significantly in the ability of trust companies to weather the Panic. The trust companies that followed a retail‐oriented strategy suffered a relatively greater loss of deposits—except for three small and young trust companies noted earlier, which probably accounted for an apparent positive effect of distance from Wall Street. Location does matter as a material determinant of consumer behavior.⁷ Thus, location warrants continued study as a proxy for attributes of the business model.

Second, depositors seemed to discriminate and discipline the bad managers, Heinze, Morse, and their associates. The deposit runs were concentrated especially on trust companies associated with Heinze and Morse whereas other trust companies experienced milder runs. The prevalence and significance of the Heinze–Morse effect predominate in the deposit changes for Manhattan trust companies.

Finally, the market power hypothesis gains weak support, at best. On one hand, Table A.2 shows that affiliations with the financial elite and with the NYCH yield insignificant correlations with deposit changes (rows 15–16, columns A–F). Affiliation with the financial elites appears in none of the stepwise regression results. Also, affiliation with the NYCH is absent from the regression results except for Panel 6 in Table A.3. This glimmer of association with deposit changes in Table A.3 offset by the absence of correlations for the NYCH dummy variable in Table A.2 yields no strong endorsement of the market power hypothesis.

An implication of these findings is that the resistance J. P. Morgan encountered as he tried to mobilize collective action among the trust companies sprang from differences in business strategy and from wariness about which firms were mired in the Heinze–Morse debacle. As the game theory model (discussed in Chapter 24) suggests, cooperation depends profoundly on trust; and trust among such diverse players must have been in short supply.

Further Research

This analysis has focused specifically on Manhattan‐headquartered trust companies. Nearly 25 trust companies resided in the outer boroughs and could be studied for similar effects. In addition, state and national banks in New York City could be added to the sample.

Measures of liquidity and capital adequacy could help to distinguish among subgroups of the trust company community. Similarly, the fact that some trust companies increased their deposits during the Panic warrants explanation: Why did they gain?

Finally, the material (and sometimes significant) cross‐correlations among independent variables cautions the interpretations of regression estimates and invites the application of other econometric techniques to strengthen the inferences about the independent variables.