The Receipt Window

An account holder who wanted to convert trade coin into bank money could sell the coin to the Bank at a posted official price, receiving in return (1) credit to his or her bank account and (2) a receipt for the coin sold. The receipt allowed a person to repurchase the coin for a small fee. A receipt was, in modern terminology, an American call option on the type of coin sold (and no other coin) with an expiration date of six months after the sale, and a repurchase, or “strike,” price slightly above the original sale price.

Take, for example, a transaction: in February 1737, Jan Albert Vos sold 800 silver ryxdaalder coins to the Bank for 1,920 bank florins and a receipt. The Bank’s account liabilities increased with a corresponding increase in assets. See Table 2.A. To the Bank, the coins were in effect “encumbered” (E) by the call option embedded in the receipt. Receipts were not deposits in the modern sense of a generalized demandable debt claim against the Bank. Instead, such sales generated a negotiable claim against only a specific type of collateral; that is, a receipt.

Table 2

THREE WAYS TO CREATE BANK MONEY WITH BALANCE SHEET EFFECTS

a To simplify accounting, the Bank adjusted the assets to a record-keeping agio of 5 percent. At sale, the reverse adjustment creates a profit.

Sources: A. Amsterdam Municipal Archives [hereafter AMA] 5077/297, folio 1308 and AMA 5077/1378, folio 47. B. AMA 5077/1378, folio 44. C. AMA 5077/297, folio 1503.

For the silver trade coins that constituted the bulk of the Bank’s receipt business, the repurchase price of a receipt was generally 1/4 percent higher than the sale price. For a few favored domestic silver coins, this margin fell to 1/8 percent, and it rose to 1/2 percent for gold coins. In the rare instances of rate adjustment, the Bank standardized these outlier rates to 1/4 percent, but such coins were never major sources of receipt business.Footnote ⁷ Receipts could be extended for another six months at the same cost as for redemption, and receipts were fully negotiable as bearer instruments.Footnote ⁸

The great majority of the time, the call option inherent in a receipt was “in the money”; that is, the price for repurchasing the coin listed on the receipt was below its market value, so the receipt was eventually redeemed. The function of a receipt was thus much like a modern central bank (term) repo, providing liquidity to the money market against liquid collateral at a policy interest rate.Footnote ⁹ From the beginning of the receipt system in 1683, the Bank’s implied policy interest rate was simply the redemption fee of 1/4 percent per six months, or about 1/2 percent annualized. Such a low rate was feasible because the Bank scrupulously maintained a complete stock of all trade coins encumbered by receipts. This rate was substantially lower than contemporary rates on relatively secure but uncollateralized instruments, such as bills of exchange (with a mean return of around 4 percent) or the debt of the province of Holland (about 2.5 percent, see ‘T Hart Reference ‘T Hart, ‘t Hart, Jonker and van Zanden1997).

The repo-style credit available through the receipt window was a key advantage of Amsterdam over other contemporary financial markets. Essentially, the existence of this facility (combined with an absence of capital controls) meant that trade coins could be more liquid in Amsterdam than elsewhere, since coins under receipt could be utilized for transactions in the Amsterdam market, pending their shipment to other markets where their highest value could be realized.Footnote ¹⁰ By this means, the receipt facility allowed Amsterdam merchants to carry long positions in trade coin at a low cost. Johann Philipp Graumann (Reference Graumann1762, p. 137), an experienced trader in coins and bills of exchange, recognized this function of the receipt facility and described it as follows (our translation):

Despite the parallels, there is an essential difference between the receipt system and modern central bank repos. The policy interest rate inherent in the receipt system was an administered rate rather than a target for market rates, as is preferred by many central banks today. The quantity of credit generated through this standing facility was not under direct control of the Bank. This, however, did not mean that the Bank was indifferent to fluctuations in its money stock.

Operations in Unencumbered Coin

An alternative way for Amsterdam merchants to acquire Bank funds was to purchase such funds from people with accounts at the Bank. Inside the Bank, florins were transferred from one account to another. Outside the Bank, coins were delivered in the opposite direction. This was typically done through dealers who were active in a secondary market that took place in front of the Bank and charged bid-ask spreads of 1/8 percent or lower.

The coin behind this secondary market was the Dutch silver gulden. It was worth one current guilder and acted as both the unit of account and the basic medium of exchange for the Dutch economy. The prices of goods, services, real estate, taxes, and financial products (other than international bills of exchange) were denominated in current guilders. Purchases could be paid with gulden coins or with private arrangements such as cashier accounts or personal IOUs that represented gulden coins. To clarify presentation, we use “florin” for bank unit of account, “guilder” for current unit of account, and “gulden” for the coin worth one guilder while noting that contemporaries used these terms interchangeably.

The word agio derives from Italian for an exchange rate premium, and in Amsterdam the agio was as the premium of bank florins relative to current guilders. By the eighteenth century, it was the universally acknowledged barometer of the florin’s monetary value. The market agio was routinely reported in the financial press and Amsterdam merchants would have been well informed about its current value. Since the metallic content of a silver gulden coin was largely constant during this time, a stable agio also implied a stable metallic value for the bank florin.

Figure 1 plots monthly values of the market agio from December 1735 to January 1792. A ratio of 1.05 current guilders per bank florin was reported as an agio of 5 percent. The agio stays between 4 and 5 percent for most of the sample except during the Seven Years’ War, and during the period of the Bank’s decline from the Fourth Anglo-Dutch War onward.

Figure 1

MARKET AGIO, 1736:1–1792:1 (Percent Premium Bank Florin over Current Guilders)

Notes: The shaded band represents the Bank’s unofficial target band. The Seven Years’ War period is taken as August 1756 (the first Prussian campaign of the war) through July 1763 (the outbreak of the postwar financial panic in Amsterdam). The Fourth Anglo-Dutch War period is taken as December 1780 (declaration of war) through May 1784 (signing of the Treaty of Paris). The series shown in the figure is augmented by one month at the end of the sample to reflect the date conventions in the archival data. Sources: Gillard (2004) and Schneider, Schwarzer, and Schnelzer (1991).

The Bank routinely intervened in the agio market by buying and selling large quantities of gulden coins at the going market price in the same way as modern central banks buy and sell securities in outright open market transactions. No receipts were granted for these purchases because gulden coins were not considered trade coins. For an example, we return to 1737. The Bank purchased 25,945.375 current guilders (in the form of gulden coins and small change) from a dealer named Arnoud Borchers at an agio of 3–15/16 percent. Borchers got 25,000 bank florins with a corresponding increase in coin held by the Bank (see Table 2.B). The Bank’s purchase operations, together with trade coins whose receipts had expired, gave rise to a stock of unencumbered coin, U, in the Bank’s vault. Coins acquired in this fashion were owned outright by the Bank and could be readily sold back into the market.

For long periods, the agio traded in a range of 4 to 5 percent. Those thresholds derived from the coinage laws of the Dutch Republic. Each domestic trade coin C had an implicit agio a_C, defined by

$$a_c = 100 \times \left( {\left( {\frac{{{\text{legal value of coin }}C{\text{ incurrent guilders}}}} {{{\text{legal value of coin }}C{\text{ in bank florins}}}}} \right) - 1} \right).$$

For example, ordinances declared each ryxdaalder coin to be worth 2.5 current guilders outside the Bank, versus 2.4 florins within the bank. Ryxdaalder coins put under receipt created an implicit agio of: $$a_{ryxdaalder} = 100\left[ {\left( {\frac{{2.5}}{{2.4}}} \right) - 1} \right] = 4.167$$. Each Dutch trade coin had a slightly different implicit agio because bank florin and current guilder values varied slightly (see Polak Reference Polak1998). These implicit agios applied to the first “leg” of the receipt agreement and set the floor of the range. These agios were implicit, however, because the bank did not exchange trade coins for current guilders at the receipt window.

To repurchase in the second “leg,” customers also had to pay the 1/4 percent fee, so the implicit agio for each coin leaving the Bank was greater than when entering by that amount. Before the 1683, Bank fees had been higher, and, consequently, the range had been wider (Quinn and Roberds Reference Quinn and Roberds2014, p. 3). The consistently low policy rate translated into the consistently narrow range exhibited in Figure 1.

The range could act as an anchor because of specie-flow incentives. For example, a market agio higher than a coin’s implicit agio encourages the inflow of those trade coins, and the resulting increase in the stock of bank florins would then push down the market agio. As with other specie-flow examples, such as the classical gold standard, these incentives can prove weak relative to liquidity values, expectations of future price movements, and the other reasons people exchange money. In the Dutch case, most of the coins placed at the Bank were foreign with guilder values at market prices. Such coins offered no assurance of arbitrage.

Even the Dutch trade coins accepted by the Bank were rarely used in regular circulation. They traded at a liquidity premium and had to be acquired for a fee from mints or money changers. For example, Graumann (Reference Graumann1749, p. 13) notes that “[d]ue to their widespread use [abroad], Dutch silver rijksdaalders trade several percent higher than their domestic value, and in the Baltic their value is bid up even higher.” And “[b]ecause the Dutch East India Company can sell Dutch silver rijders profitably in the East Indies, they trade at several percent above their Bank price” (Graumann Reference Graumann1749, pp. 13–14). As a result, the market agio could trade outside the implicit target range, especially if few domestic trade coins were under receipt at the Bank.

Also, receipts distorted specie-flow incentives. A market agio below an implicit agio encouraged redemptions, but one had to have a receipt to take advantage. Receipts could be purchased from other people, so Pit Dehing (Reference Dehing2012, pp. 124–26) argues that receipt prices had an inverse relationship to the market agio. While there is insufficient data to confirm the strength of this relationship, paying more to acquire a receipt reduces the arbitrage gains. Similarly, the ability to sell a valuable receipt undermines the disincentives to bring trade coins to the Bank when the agio is low. In contrast, receipts have little value when the agio is above the target range, so incentives are less affected. In the extreme, receipts can have so little value that people do not pay to roll them over. As a result, specie-flow incentives are stronger for agios above the target range than below. In Figure 1, the agio sometimes lingers below the anchor for extended periods. In no period did the agio linger above the anchor.

Unsecured Loans

Yet another way to create bank accounts was for the Bank to engage in unsecured lending. The great bulk of such loans were made to two privileged borrowers, the Dutch East India Company (“Company,” also known by its Dutch initials VOC) and the City.

The Company frequently borrowed balances from the Bank against unsecured notes known as anticipations, which were to be repaid by the sale of goods in transit from Asia to the Netherlands. Loans to the Company show up in the Bank’s balance sheet as increases in loan assets (L) and increases in account liabilities M. For example, a loan of 100,000 florins to the Company increased the stock of bank money by the same amount (see Table 2.C). Such short-term borrowing allowed the Company to outfit one year’s trading fleet and pay out dividends to its stockholders, while awaiting the return of a previous year’s fleet. These loans also helped the Company purchase the large stocks of precious metals that were carried by its fleets to Asia (Gaastra Reference Gaastra and Richards1983). The city council authorized the Company’s line of credit, and the interest earned was an important source of income to the Bank (Uittenbogaard Reference Uittenbogaard and van Nieuwkerk2009). Most of these loans appear to have been granted automatically at the discretion of the Company.

Since the City owned the Bank, its status as a borrower was different. Until the 1780s, loans extended to the City carried no interest and were operationalized through the removal of unencumbered coin U from the Bank’s vault. These had no direct effect on the amount of bank florins outstanding. The City attended to these “loans” when and how (write-down or repay) it wanted, so they functioned as adjustments to the Bank’s equity (ε). This also occurred when the City took the Bank’s residual profits as seigniorage.

City loans enter account transactions only when the City began to borrow bank balances in 1782. The City directed most of these new bank florins to a new municipal lending agency (Stadsbeleeningkamer or “municipal loan chamber”) that the City used to disperse loans to individuals. The loan chamber did reliably repay the bank with 2 percent interest, but the chamber continuously re-borrowed to do so. It is unclear whether the Bank was permitted to ration the supply of credit to the chamber.

Measuring the Three Tools

To measure the three monetary tools, we reconstructed each transaction that altered the amount of bank florins from January 1736 through December 1791 (Quinn and Roberds Reference Quinn and Roberds2019). Records of these transactions exist because the City of Amsterdam maintains the ledgers in its municipal archives. We begin in 1736 when the Bank simplified its internal accounting processes, and end with the last year with complete records. The Bank maintained meticulous double-entry records, so the Bank’s master account contains the relevant transactions, a total of 73,479 entries, or 1,336 transactions per year on average.Footnote ¹³

Figure 2 gives the total level of account balances for 672 months. In terms of the balance sheet (Table 1), this is the level of the Bank’s total monetary liabilities M, and that level is stable over long periods of time. The total stays between 15 and 25 million florins for 88 percent of the months. The only major deviation is the peak surrounding the crisis of 1763 (Schnabel and Shin Reference Schnabel and Shin2004). This stability is in no small part explained by active policy on the part of the Bank. To see that, Figure 2 disaggregates the monetary total into encumbered, unencumbered, and loans.

Figure 2

MONETARY LIABILITIES BY BACKING ASSET, 1736–1791

Note: For seven months of the sample, ML is slightly negative. Sources: AMA and authors’ calculations.

The Bank’s account ledgers do not detail whether a change in Bank balances stems from a receipt/redemption (a change in E), a loan/repayment (a change in L), a purchase/sale (a change in U), or an adjustment in equity (ε) such as from fees and interest payments. That information mostly resides in another set of books that records flows of metal and related fees.Footnote ¹⁴ Referring again to Bank’s balance sheet in Table 1, these “cash books” track changes in coins on the asset side of the balance sheet. Through the arduous reconciliation of the two sets, we identify an offsetting change in the balance sheet for each change in monetary liabilities.Footnote ¹⁵ Using this information, we can separate monetary liabilities M by the three other constituent parts of the balance sheet through the identity M = M_E + M_U + M_L. The component M_E is by definition equal to E, encumbered coin in the Bank’s vault. M_U is equal to U, unencumbered coin owned outright by the Bank. M_L is L, lending by the Bank primarily to the Dutch East India Company, less ε, the City’s various takings of Bank equity.Footnote ¹⁶

This deconstruction reveals considerable movement in the three monetary components. The most striking result is the dramatic variation in the level of money backed by coins under receipt M_E. For example, the 1764 peak in total monetary liabilities of 32 million is almost completely backed by receipts. At the other extreme, M_E is only 375,000 in 1783. Our period opens with receipts accounting for two-thirds of all bank florins. Eight years later, that is down to one-fifth. At its peak, receipts account for 97 percent of all bank money. At its nadir, encumbered accounts are only 2 percent.

To maintain the stability of its monetary total, the Bank adjusts. The Bank uses its own metal to change monetary levels, and these adjustments cause the level of monetary liabilities backed by unencumbered metal (M_U) to range from a high of 12 million (60 percent of bank florins) in 1751 to a low of 0.6 million (2.5 percent) in 1763. Purchased metal seems to offset low receipt levels. In contrast, when receipt levels were strong, the Bank ran out of metal to sell. The Seven Years’ War appears an extreme example.

The third component is money backed by loans less equity (M_L). The series exhibits strong seasonal variance, but it has more long-term stability than do the other two series. Until 1760, the Dutch East India Company routinely owes the Bank a few million bank florins in revolving credit. Then, for two decades, the Company repays most advances within a year. Starting 1780, lending to the Company and then to the City grows quickly because of the Fourth Anglo-Dutch War. This regime change coincides with a collapse in receipt funding, default by the Company, and the permanent decline of the agio (Quinn and Roberds Reference Quinn and Roberds2016).

Figure 3 reorganizes the data around the level of receipts M_E to emphasize findings relevant to our characterization of the Bank’s policy framework. The horizontal axis displays M_E while the vertical axis measures the sum of M_U and M_L. The horizontal and vertical distance of each observation sums to the total amount of bank florins at that time. Figure 3 illustrates how thoroughly the Bank stabilized the total money supply. For a linear fit, a slope coefficient of negative one would mean the total money supply on average was unchanging as receipt levels varied. For the observations in Figure 3, the estimated slope for a regression of M_E on (M_U + M_L) is −1.004. The estimated intercept also suggests a target quantity of bank florins of 19 million. Such consistency over decades of observations demonstrates impressive stabilization in the face of substantial swings in receipt volumes. The regression analysis in the next section will investigate which of these components responded to keep the total so stable.

Figure 3

MONETARY COMPONENTS BY MONTH, 1736–1791

Sources: AMA and authors’ calculations.

The extremely high receipt observations from the end of Seven Years War (squares in Figure 3) show that stabilization of the quantity of money was secondary to accessibility. At that time, demand for receipts was booming as very large amounts of metal sought the Bank. After selling what it could (M_U), the Bank let the level of bank money break pattern rather than adjust the terms of lending to discourage receipts. In terms of the framework, the goal of a maximum receipts was well satisfied even if the target of quantity stabilization was not. The same applied to the agio. At the end of the Seven Years War, the agio traded well below 4 percent, but the Bank did not respond by tightening receipt terms.

The other telling extreme is the period that we know was covered by the instruction of 1782. During the 1780s (triangles in Figure 3), the level of receipts was low and the agio was in decline (Figure 1). The Bank had more than 3 million in M_U that it did not sell. Stephen Quinn and William Roberds (Reference Quinn and Roberds2016, p. 64) argued the Bank resisted because a sales operation might not have been effective. The Bank could neither compel the Company to service its loans nor compel the City to recapitalize the Bank, so loans gone bad undermined the agio more than the available quantity adjustment could correct. The more comprehensive perspective of this article suggests a simpler explanation: quantity stabilization took precedent over agio stabilization. The Bank did not sell in 1780s (triangles in Figure 3) when such action would have reduced the total stock of bank florins to the neighborhood of 13 million. Extreme periods, however, do not reveal how the Bank handled agio targeting during the decades of relative stability that comprise most of the sample.

Table 4

OLS REGRESSION OF M_E ON M_U + M_L

Note: Standard errors apply the Newey-West (1987) correction for heteroskedacity and serial correlation up to 13 months.

Sources: AMA and authors’ calculations.

Measuring Responses over Time

To see how the three components of the money supply responded to stabilize the quantity of money and how the Bank responded to the agio, we fit a VAR model to the data series. The VAR approach estimates how a movement in one variable alters the later behavior of every variable. The approach delivers a measure of response over time, and it can indicate whether an increase in the level of receipts (M_E) leads the Bank to eventually sell, and whether an increase in the agio leads the Bank to eventually buy.

The estimated VAR includes five variables. These are the three components of Bank money balances: M_E, M_L, and M_U, plus the market agio, and bill rates.Footnote ¹⁷ Bill rates represent the attractiveness of holding bank florins, and these are calculated as the projected annualized return on bills of exchange circulating between Amsterdam and London.Footnote ¹⁸ The VAR was left unconstrained and was estimated by OLS under the usual diffuse-prior interpretation of such estimates. The financial market variables (agio, bill rate) come first in the (Choleski) orderings, the intuition being that these would react quickly to changes in international conditions and market sentiment. The monetary components come second, the intuition being that these could be somewhat slower to react than market prices. Estimated impulse responses are robust to changes in orderings within the two classes of variables.Footnote ¹⁹

How did the Bank react to receipts? Figure 4 shows the response of Bank operations M_U from a change (an impulse) in M_E over time. The initial negative value means that an increase in receipts of 670,000 bank florins (one standard deviation) led to the sale of 120,000 bank florins the next month, and the persistence of the negative result means that the response endured. The result is consistent with the Bank having a policy of using operations to counter receipts. In contrast, the second part of Figure 4 shows the response of receipts to a one standard deviation operation (311,000 bank florins). The initial response was positive, but the effect dissipated. From the perspective of a year later, operations responded negatively to receipts, but receipts did not respond to operations.

Figure 4

RESPONSES TO MONETARY IMPULSES (By Month in Millions of Bank Florins with 70 Percent Error Bands)

Source: Authors’ calculations. See the online supplemental appendix for the full impulse-response matrix.

And what about the agio? Figure 5 shows that a change in the agio did encourage operations to stabilize the agio, but a jump in the agio did not send the Bank into action the next month. Instead, the Bank tended to respond if the agio persisted for a few months. That timing, however, corresponds with how receipts were negatively reacting to the agio, in the second part of Figure 5. Figure 4 already shows that operations responded to receipts and not vice-versa, so Figure 5 can be read as another version of the sterilization story. A one-standard-deviation upward shock to the agio (29 basis points) eventually induced a 400,000-florin decline in M_E, which was partially offset by a 160,000 increase in M_U.

Figure 5

RESPONSES TO AGIO IMPULSES (By Month with 70 Percent Error Bands)

Source: Authors’ calculations. See the online supplemental appendix for the full impulse-response matrix.

The somewhat ambiguous verdict on agio stabilization might follow from the years when the agio was well below target. To ask the same question of more ordinary months, Figure 6 excludes September 1756 to October 1763 and observations after 1780. The Bank still responded to the agio with operations (first part of Figure 6), but the responsiveness of receipts to the agio is greatly diminished (second part), so the result is stronger support that the Bank was using operations in a manner consistent with agio stabilization.

Figure 6

RESPONSES TO AGIO IMPULSES BY “ORDINARY” MONTHS: 1736:1 TO 1756:8 AND 1763:11 TO 1780:12 (With 70 Percent Error Bands)

Source: Authors’ calculations. See the online supplemental appendix for the full impulse-response matrix.

The VAR approach also measures whether the agio returned by itself to average levels, rather than from changes in the monetary components. The last part of Figure 5 reports the response of the agio to itself for the entire sample. The agio was auto-regressive with a half-life of 12 months. A year after a 30 basis-point jump, the agio was still 15 basis points above average. The last part of Figure 6 finds that the agio was much more auto-regressive during ordinary times. The half-life falls to three months. This reversion to average supports the contention that the agio market used expectations to auto-stabilize, especially when war was not stressing the system.

What role did loans play? The VAR results (not shown here) find that new credit led to declines in the other two series, but not vice versa.Footnote ²⁰ That supports the conclusions that the Company was choosing borrowing levels and that the Bank countered directly, if modestly: the VAR estimates that the Bank offset about 15 percent of upward shocks to credit through unencumbered coin sales. A one-standard deviation credit shock of 320,000 florins leads to 49,000 florins in sales. The greater response, however, was through receipts, and that took months to develop. Only a persistent change in credit produced an offsetting change in receipts. Finding that customers eventually adjusted to unsecured lending confirms that tension existed between the goals of profit creation and receipt demand.

In summary, the VAR approach finds that the Bank used operations to sterilize receipts and that the agio auto-corrected, especially during ordinary periods. There is also evidence that the Bank responded to the agio with operations, but the relationship is murky because the agio also effected receipts. The VAR, however, is ill suited to test for the non-linearity of operations springing into action when the agio escaped its anchor range. To isolate the response of operations to thresholds, the next section takes a different approach.

Testing for Constraints

The previous section asks how long it took the Bank to respond. This section instead asks what environment coincided with a response. This change concedes explanatory power to gain a model better suited for the testing of non-linear constraints. For example, did the Bank respond when the agio crossed the five or four percent thresholds as suggested by the instruction of 1782? Another asymmetry to model is that the Bank can only sell if it has sufficient coin (M_U), but it can always buy. Such a constraint would apply to both quantity stabilization and agio stabilization, so interacting the M_U level with extreme agios can test for just the constraint applying to low agios. Table 5 summarizes the expected effect of these constraints on the probability of the Bank beginning an operation to sell or to buy.

Table 5

EXPECTED SIGNS IF OPERATIONS WERE USED TO STABILIZE THE AGIO

Source: See text.

To conduct the test, we fit a discrete choice (multinomial logit) model. Each month of the sample is assigned to one of three regimes: tightening, easing, or no intervention. A model then estimates the probability of transitioning from no intervention into either tightening or easing during the following month as a function of variables observed during the current month.Footnote ²¹ Because these models tend to be weakly identified, only sparse specifications were estimated. Explanatory variables are the current month’s regime (a logit model is estimated for each regime), the five series from the previous section (M_E, M_L, M_U, agio, and bill rate) and the variables in Table 5: dummy variables for a high agio and for a low agio, and two variables that interact those dummy variables with the level of coin owned outright. The purpose is to see (1) if the instruction of 1782 was followed, and (2) if a lack of coin impaired the Bank from selling when it otherwise would have.

To perform this analysis, we must define a move to tightening and to easing. As a first step, we partitioned the sample into three observable regimes, based on the sign and magnitude of the Bank’s net interventions over a given month: (1) a “draining liquidity” regime (consisting of months where the Bank substantially reduced the stock of bank money through unencumbered metal sales), (2) an “adding” regime (months with substantial net unencumbered metal purchases), and (3) a regime of “no intervention” (months with neither draining nor adding). Draining (adding) was defined as a monthly rate of open market sales (purchases) in excess of 25,000 florins. The 25,000-florin filter was chosen so as to screen out months where only maintenance-type open market transactions occur, the idea being that larger operations were more likely to result from purposeful policy interventions. The sample Markov transition matrix for the three regimes (draining, adding, no intervention) is given by

$$\pi = \left[ {\matrix{ {.525} & {.0667} & {.408} \cr {.057} & {.631} & {.312} \cr {.119} & {.107} & {.774} \cr } } \right],$$

where π(i,j) gives the probability of transition from regime i to regime j. The corresponding steady-state distribution over regimes is (.179, .210, .611). In words, the Bank intervened heavily during about 40 percent of our sample (261 out of 672 months), with interventions roughly evenly split between sales and purchases. Regimes were moderately persistent, and it was rare for the Bank to go directly from adding to draining liquidity, or vice versa.

The first result was that a “high” agio of 5 percent or more lacked sufficient observations for meaningful estimation. This is strong evidence that the market discipline rather than the Bank operations kept the agio below this upper threshold. To achieve identification, the model reported in Table 6 instead uses a “high” agio of 4.8 percent, which is the 75^th percentile of the agios in our sample. The first version uses a “low” agio of 4 percent as per the instruction of 1782. The second version defines a low agio of 3.7 percent (25^th percentile). This mirrors the definition of the high agio (75^th percentile) and creates a low agio with timing nearly equivalent to the Seven Years War and the post-1782 eras (see Figure 1). The second version’s low agio corresponds with the two extraordinary periods when the agio anchor failed.

Table 6

MULTINOMIAL LOGIT MODELS OF TRANSITION PROBABILITIES: POSTERIOR MEANS AND “T-STATISTICS” OF COEFFICIENTS

* = 90 percent credible interval (“Bayesian confidence interval”) for that coefficient not containing zero.

** = 95 percent credible interval (“Bayesian confidence interval”) for that coefficient not containing zero.

Notes: Estimates in the table were calculated with the BRMS package in the R programing language using diffuse priors. Each estimate uses 20,000 draws, sampling every tenth draw. T-statistics are posterior means divided by posterior standard deviations.

Sources: AMA and authors’ calculations.

Table 6 gives results for moving from no intervention to either draining or adding. The results do not support the contention that the Bank mechanically followed a threshold-style policy as described in Table 5. The agio by itself has the wrong sign on transitions to sales (0.214) and to purchases (−0.455). A low agio does not have a positive coefficient on selling (−1.175), and a high agio does not have a positive one on buying (−0.237). An agio story that does emerge has the Bank not buying when the agio is low (−4.475**) and not selling when high (−1.915), but avoidance of unprofitable and destabilizing operations is not the active implementation suggested by the instruction of 1782. The second version of the model, which further restricts a low agio, does not change these results (see the table’s lower panel).

Although the Bank did not start selling just because the agio was low, it did tilt towards a response if the stock of owned metal was substantial enough (0.367). The coefficient suggests that it took 3.2 million bank florins in owned metal (M_U) to offset a low agio (0.367*3.2 = 1.175). The average stock of owned metal over our sample is 4.9 million with a standard deviation of 2.8 million, so it took a substantial amount to tip the Bank towards responding to a low agio with a sales operation. This coefficient grows larger and slips into statistical significance (0.569*) in the second model.

In summary, the Bank was not initiating operations in response to the agio crossing upper or lower thresholds. Instead, high agios corrected themselves, and low agios recommended sales over purchases only when the Bank owned plenty of metal. The focus on levels of owned metal also expressed as a powerful urge to purchase as the level of M_U fell (−0.368**). The combination of drainages from seigniorage, quantity stabilization, and the difficulties of bolstering low agios made rebuilding unencumbered assets a priority.