Content uploaded by James Stodder

Author content

All content in this area was uploaded by James Stodder on Jul 18, 2016

Content may be subject to copyright.

The Macro-Stability of Swiss WIR-Bank Credits: Balance, Velocity and Leverage

1

James Stodder, Rensselaer Polytechnic Institute, Hartford, CT, USA; jim.stodder@gmail.com, (860) 690-0213

Bernard Lietaer, Center for Sustainable Resources, Univ. of California-Berkeley, bernard@lietaer.com

Abstract: Since 1934 the Swiss Wirtschaftsring/Cercle Économique (Economic Circle), now the WIR-Bank,

has issued its own currency, not backed by Swiss Francs. Turnover in WIR is countercyclical: firms use it more

in a recession. A money-in-the-production-function (MIPF) model implies that this spending arises through

larger WIR Balances for larger firms, but greater WIR Velocity for smaller ones. Panel data by industrial sector

confirms this pattern, similar to commercial trade credits, a major source of non-bank credit. The

countercyclical multiplier on WIR expenditures is highly leveraged, and its impact on the Swiss Economy

greater than its turnover would suggest. Keywords: complementary or community currency, countercyclical

JEL Codes: E51, G21, P13.

I. Introduction

The Swiss Wirtschaftsring or “Economic Circle,” founded in 1934, is nowadays called the WIR-bank.

Those studying reciprocal payment mechanisms refer to it as a “social,” “community,” or “complementary”

currency – terms that are broadly equivalent. The WIR is really a centralized credit system for multilateral

exchange, not a physical currency per se.

Despite its origins in the ideas of Silvio Gesell – a monetary economist praised by Keynes (1936) – the

WIR has attracted little attention from economists. Studer (1998) and Stodder (2009) provide the only formal

empirical studies. Studer (1998) shows that WIR credits are positively correlated with the growth of the Swiss

money supply. Stodder (2009) shows that WIR bank transactions are also highly countercyclical –more so than

the official Swiss money supply itself. If a secondary currency provides added financial stability, then standard

monetary policy may not be optimal.

The idea that community currency expenditures will be countercyclical has a long currency. It is

accepted by Yale’s Irving Fisher in his Stamp Scrip (1933), a short book that documents the flourishing of such

currencies in the US of the Great Depression. Building on Stodder’s (2009) empirical demonstration of

countercyclical WIR spending, our present paper explains the different commercial motivations for large and

small firms – the latter with more restricted credit access – in their use of WIR. We can thus show how these

differing motivations create a credit interaction that makes WIR Turnover (or total expenditure) highly

countercyclical.

1

Article is published in Comparative Economic Systems, 2016, http://link.springer.com/article/10.1057/s41294-016-0001-5. We

wish to thank Stefan Winkler, a statistician for WIR-Bank, for his generous aid in providing and interpreting WIR-Bank sectoral data.

Thanks also to Peter Pedroni, David Giles, discussants at the Centre for European Research in Microfinance, University of Brussels,

and participants at the First International Conference on Community and Complementary Currencies, University of Lyon. They bear

no responsibility for any remaining errors.

2

Stodder’s earlier (2009) paper considers the countercyclical pattern of WIR Turnover (Turnover = WIR

Balances times Velocity), but lacks data to distinguish Balances from Velocity. With a new disaggregated data

set, we can now show that WIR Balances and Velocity are both countercyclical drivers, but for different types

of businesses. Larger Non-Registered (i.e., non-member) firms – free to accept as much or as little WIR-

currency as they wish – accept more WIR when other forms of money are in short supply, in a recession. For

such Non-Registered firms, WIR-Balances are the countercyclical term. Smaller Registered firms, by contrast,

will be shown to have countercyclical WIR-Velocities. Thus both Registered and Non-Registered firms show a

countercyclical Turnover, but dominated by different terms. This pattern for WIR between large and small

firms is argued to be highly analogous to Trade Credits, a major form of non-bank credit in developed

economies, also countercyclical (Nilsen, 2002).

There are hundreds of community currency systems in existence today, described in a descriptive

literature, largely by non-economists (Williams, 1996; Greco, 2001; Gomez, 2008). The Swiss WIR-Bank is

the largest such system, with over 70,000 customers throughout the country. Formal membership and

registration is restricted to Small and Medium Enterprises (SMEs) (Studer, 1998).

The Swiss WIR-Bank, is the largest and oldest surviving ‘club’ form of money of which we know. The

finding by Stodder (2009) that WIR activity is countercyclical is based on data from 1948 to 2003. Our present

study with more recent data (i) strengthens that conclusion, (ii) performs further tests for structural breaks, and

(iii) shows how large and small firms interact to structure this countercyclical resilience.

The paper is organized as follows. Section II tests the countercyclical record on a long time-series of

WIR data. Section III examines WIR’s structure of smaller and larger clients. Section IV presents some basic

theory on how firms will use secondary or ‘residual’ currency. Section V runs panel regressions to test this

theory. (‘Bootstrap’ simulations are first required to build up a large sample for cointegration testing.) Section

VI breaks our panel into sectors and uses Chi-Squared tests to summarize 192 separate regressions. Section VII

summarizes our results and considers the implications.

3

II. WIR’s Countercyclical Record: Time-Series Regressions

We first turn to our basic countercyclical result. Stodder (2009) uses Vector Error Correction (VEC)

models to check the stability and cyclical nature of WIR expenditures. If WIR grows with the economy, then

the long-term relationship between them – shown by the coefficient on GDP in the Error Correction (EC)

portion of the VEC – should be positive. But if WIR activity is also countercyclical, then the correlation

between short-term changes in GDP on the one hand, and changes in WIR activity on the other, should be

negative – as shown in the Vector Auto Regression (VAR) portion of the VEC. These results contrast long-

term “secular” growth with short to medium-term “cyclical” deviations.

Stodder (2009) uses VEC models, so the non-stationarity of all variables is a necessary condition for

cointegration. Standard Augmented Dickey-Fuller (ADF) tests show all variables to be integrated of order 1, or

I(1). Integration is not tested for breaks in constant or trend in this earlier paper, however. The faltering trend

of WIR growth post-1992 (see Figure 1) does suggest such breaks, so tests are useful. The present paper

implements structural break tests and furthermore, uses ARDL models to replicate the results of VEC models

when both are feasible; i.e., when all endogenous variables are I(1). Our key variable, WIR Turnover, is shown

countercyclical in both types of cointegrated regression.

Figure 1 shows WIR Turnover and Swiss Imports as a percent of GDP from 1948 to 2013, with

recession bars shown up to 2014. There is a sharp dip in Imports for most recessions. WIR Turnover is seen to

peak in the early 1990s, at the end of a long recession and a trend of falling Imports. The steady decline in

Turnover since 1992 suggests that WIR activity may be negatively correlated with the percent of Swiss GDP

going to imports. This is intuitive, since WIR currency is normally accepted by Swiss firms only, and a subset

of Swiss firms at that. Greater internationalization of the Swiss economy therefore works against WIR activity

– and limits it even more to the Small and Medium Enterprises (SMEs), where, as will be seen, it is already in

greatest use.

4

Figure 1: Real WIR Turnover, Imports as % of GDP (1948 to 2013)

Sources: Turnover from WIR Annual Reports, Swiss Imports, GDP, and GDP-Deflator from State Secretariat

for Economic Affairs (http://www.seco.ch); Recessions 1962-2014 shown by OECD and Federal Reserve

(https://research.stlouisfed.org/fred2/series/CHEREC#) as in recession for at least six months. Years were

marked recessionary from 1948 to 1961 if real GDP grew by less than 1 percent.

Table 1 shows the names of variables we will use in the empirical estimates to follow.

Table 1: Notation for following Tables

1

LrTURN(-t)

Log of Real WIR TURNOVER, lagged -t period(s)

2

LrTURNa(-t)

Log of Real WIR TURNOVER, averaged t and t-1, lagged -t period(s)

3

LrGDP(-t)

Log of Real GDP, lagged -t period(s)

4

LrGDPa(-t)

Log of Real GDP, averaged t and t-1, lagged -t period(s)

5

LrIMP(-t)

Log of Real IMPORTS, lagged -t period(s)

6

LrIMPa(-t)

Log of Real IMPORTS, averaged t and t-1, lagged -t period(s)

7

LrVA(-t)

Log of Real VALUE-ADDED, lagged -t period(s)

8

LrVAa(-t)

Log of Real VALUE-ADDED, averaged t and t-1, lagged -t period(s)

9

LrBAL(-t)

Log of Real WIR BALANCES, lagged -t period(s)

10

LrBALa(-t)

Log of Real WIR BALANCES, averaged t and t-1, lagged -t period(s)

11

LrVEL(-t)

Log of Real WIR VELOCITIES, lagged -t period(s)

12

LrVELa(-t)

Log of Real WIR VELOCITIES, averaged t and t-1, lagged -t period(s)

13

CointEqRes(-1)

Residual of the Cointegrating Equation, lagged 1 period

14

D( )

First Difference of any variable

All our estimates – both ARDL (Auto-Regressive Distributed Lag) and VEC (Vector Error-Correction)

regressions of WIR Turnover against GDP and Imports – face a problem of simultaneity. This arises from using

both Imports and GDP as “independent” variables. Imports are highly correlated with GDP, and modeled as

such in standard macroeconomic treatments. From Figure 1 it is clear that while both Imports and GDP fell in

5

recessions, Imports fell by more, i.e., were more pro-cyclical. Yet we need to consider Imports, since they

make WIR currency less useful.

GDP and Imports are likely to be exogenous for WIR-Turnover, but mutually determining for each

other. This is what our regressions of Turnover on both variables show: both variables are significant in

isolation, but insignificant when used together – although their Adjusted R-Square has increased. This evidence

of collinearity suggests either simultaneous, or principal component methods (Johnson and Whitern, 1992). The

latter is attractive since we are not so much interested in the relation between GDP and Imports as in their joint

effect on Turnover. Thus in Groups 3 and 4 in Table 2 below, we replace the two collinear explanatory

variables, GDP and Imports, by two principal component series, GdpImp1 and GdpImp2, that are orthogonal,

but with joint information content nearly equivalent to the original series.

Table 2: Toda-Yamamoto (1995) Tests of Granger Precedence

Group (1)

Group (2)

Group (3)

Group (4)

Lags = df = 2

Exog: 1 Lag, No Trend,

No Intercept

Lags = df = 5

Lags = df = 2

Lags = df = 4

Exog: 1 Lag, Trend,

Intercept Break

Exog: 1 Lag, Trend,

Intercept (no break)

Exog: 1 Lag, Trend,

Intercept Break

Dependent: LrTURN

Dependent: LrTURNa

Dependent: LrTURN

Dependent: LrTURNa

Excluded:

P-val.

Excluded

P-val.

Excluded

P-val.

Excluded

P-val.

LrGDP

0.2318

LrGDPa

0.7766

LrGdpImp1

0.0133

LrGdpImp1a

0.0097

LrIMP

0.7104

LrIMPa

0.9542

LrGdpImp2

0.0311

LrGdpImp2a

0.0260

Both

0.0431

Both

0.1495

Both

0.0063

Both

7.66E-05

Dependent: LrGDP

Dependent: LrGDPa

Dependent: LrGdpImp1

Dependent: LrGdpImp1

Excluded:

P-val.

Excluded

P-val.

Excluded

P-val.

Excluded

P-val.

LrTURN

0.8350

LrTURN

a

0.0220

LrTURN

0.4579

LrTURNa

0.0617

LrIMP

0.0021

LrIMPa

6.59E-04

LrGdpImp2

0.0325

LrGdpImp2a

0.0096

Both

0.0061

Both

1.12E-04

Both

0.1263

Both

0.0109

Dependent: LrIMP

Dependent: LrIMPa

Dependent: LrGdpImp2

Dependent: LrGdpImp2

Excluded:

P-val.

Excluded

P-val.

Excluded

P-val.

Excluded

P-val.

LrTURN

0.3176

LrTURN

a

0.0159

LrTURN

0.2331

LrTURNa

0.0272

LrGDP

0.0818

LrGDPa

0.0070

LrGdpImp1

0.1262

LrGdpImp1a

0.0131

Both

0.1918

Both

1.14E-04

Both

0.1561

Both

0.0013

We show exogeneity by tests of Granger precedence. Here we follow the terminology of Maddala and

Kim (1998, p.188), who note that the Granger (1988) tests show only that variation in one variable reliably

precedes, not ‘causes’ that in another. Table 2 shows Granger precedence tests for a cointegrated series,

6

following Toda-Yamamoto (1995). This test specifies the maximal number of lags suggested by information

criteria (Lütkepohl, 1991) and adds additional lags equal to the highest order of integration: here I(1). We also

add exogenous trend and intercept terms, with possible breaks on the latter. Evidence for these exogenous

terms will be given later, in unit root tests.

Note in Table 2 that the p-value for the null of no Granger precedence on the ‘Both’ term for GDP and

Imports is usually significant. This ‘Both’ p-value is also usually more significant than the product of its

component p-values; e.g., the value in row (c) is less than the product of (a) and (b) in all Groups but (3). This

suggests a joint rather than independent effect. Note also that when GDP or Imports is the Dependent variable,

the p-value on Turnover is always less significant than the remaining term; i.e. the values are always less than in

row (e) than (d); and less in (h) than (g). The p-value on Turnover is always insignificant, while that on the

other term is always significant. This shows GDP and Imports are exogenous to Turnover, and that they are

mutually determining, as previously argued.

To carry out the regressions in Table 3 using cointegration, we must first determine the integration order

of each variable. For VEC models, endogenous variables must all be I(1); for ARDL models, they may be

either I(0) or I(1), but not I(2). In order to allow for both VEC and ARDL testing, we limit our estimates to

trend, constant, and breakpoint combinations where each variable can be shown to be I(1). For the variables

LrIMPa, and LrTURNa, applying ADF tests with intercept, we can reject the null hypothesis of unit root in

levels at p-values of only at 0.091 and 0.501, respectively. That same null was rejected only at 0.121 and 0.239

for the fitted forms of these same variables. For LrGDPa, applying the ADF innovation breakpoint test with

intercept and trend, we can reject the unit root null only at 0.905. The same null is decisively rejected, however,

for first differences of all variables, with the same assumptions on intercept, trend, or breakpoint. Lag length

was based on the Schwarz criterion, and breakpoints chosen by Dickey-Fuller minimum t-statistic. Thus all

variables appear I(1) in this form, and we can use both ARDL and VEC models. For all the first-differenced

forms, however, the unit root null was decisively rejected.

7

To concentrate on the independent variables of interest – the first-differenced one-lagged terms for GDP

and Imports – other lags are not shown in Table 3. The number of lags for each variable is given in the column

heading, however. Most of the terms not reproduced were quite insignificant.

Table 3: Changes in Turnover in the WIR Exchange Network, as Explained by Changes in

GDP and Imports 1948-2013 (N=66), Moving Averages of Levels, With Up To 5 Lags

t-statistics in [ ]; ***: p-val < 0.01, ** : p-val < 0.05, *: p-val <0.10

Column:

(1a)

(1b)

(2a)

(2b)

(3a)♦

(3b)♦

Regression Form (Lags):

ARDL(3,3)

VEC(5)

ARDL(3,0,2)

VEC(3)

ARDL(3,0,3)

VEC(5)

Cointegrating Equation:

Dependent Variable: LrTURNa(-1)

LrGDPa(-1)

1.365

1.185

-23.44

-0.531

-3.327

12.030

[1.63]

***[2.53]

**[ -2.11]

[-0.32]

[-0.69]

***[5.75]

LrIMPa(-1)

25.44

7.329

8.749

-15.98

[1.39]

**[2.39]

[0.98]

***[-4.04]

Post_81

1.836

2.662

1.995

2.524

2.441

2.972

***[5.15]

***[11.77]

***[4.11]

***[8.40]

***[3.42]

***[13.67]

Trend

-0.042

-0.066

-0.051

-0.108

-0.069

-0.103

**[-2.19]

***[-5.65]

*[-1.93]

***[-6.07]

[-1.54]

***[-6.80]

Constant

-7.396

-20.25

8.749

64.768

Independent Variables:

Dependent Variable: D(LrTURNa(-1))

CointEqRes

-0.0573

-0.0454

-0.0487

-0.0366

-0.0384

0.0120

***[-5.16]

***-2.57]

***[-5.72]

***[-3.18]

***[-4.61]

[0.64]

D(LrGDPa(-1))

-0.5806

-1.151

-0.5150

-0.4288

-0.6191

**[-2.02]

***[-4.22]

[-0.86]

***[-3.36]

*[-1.83]

D(LrIMPa(-1))

-0.2738

-0.2461

0.1921

**[-2.08]

[-0.73]

[0.27]

R-squared

0.971

0.955

0.970

0.970

0.957

0.970

Adjusted R-squared

0.967

0.937

0.966

0.961

0.950

0.961

F-statistic

***147.182

***55.199

***251.291

***115.761

***136.341

***115.761

Log likelihood

255.586

149.737

146.672

151.708

141.734

151.708

Akaike AIC

-4.490

-4.500

-4.473

-4.515

-4.257

-4.515

Schwarz SC

-4.215

-3.901

-4.199

-4.031

-4.453

-4.031

(a) Bounds Test (p) (*)

< 0.01

< 0.01

<0.05

(b) Johansen Tests (p) (*)

0.0063

0.0259

4.649e-07

(c) Serial LM (p) (*)

0.2233

0.5196

0.2380

0.1582

0.1867

0.5355

Notes: ARDL estimates are via the Schwarz Criteria (SC). (*) P-values in rows (a-c) are for null hypotheses of:

(a) no long run relationship between variables, (b) no cointegration, and (c) serial-correlation does not exist. The

p-value in (a) is for the ARDL Bounds Test; that in (b) is for the Johansen Stability trace test; (c) is for the

Lagrange Multiplier test on the number of lags shown in that column.

♦ - Fitted 2SLS forms of LrGDPa and LrIMPa used in columns 3a and 3b.

If there is no coefficient shown for a term named in Table 3, then it did not appear in the cointegrated

form of the regression – D(LrGDPa(-1)) in column (2a), for example. Note that Turnover is treated here as a

dependent variable. This follows from our Granger tests of Table 2, which show GDP and Imports exogenous

to Turnover. Cointegration for all forms is confirmed by the Bounds and Johansen tests in rows (a) and (b),

8

implying (but not implied by) the Granger precedence shown in Table 2. The important structural result here is

that the first lags of GDP or Imports, here in bold, are generally both negative and significant – implying

counter-cyclicity. The exception is in column 2a. Here simultaneity biases the result, with both GDP and

Imports appearing as right-hand-side variables.

Columns (3a) and (3b) address this simultaneity problem with fitted forms of LrGDPa and LrIMPa, as in

a Two-Stage-Least-Squares (2SLS) estimate. Our first-stage OLS regressions use Exchange Rate and Total

Factor Productivity as identifying exogenous variables for LrIMPa and LrGDPa, respectively. Serial correlation

and multi-collinearity are irrelevant here; all we care about is fit, and all R2 are at 99 percent. All independent

variables are significant; results available on request.

It is encouraging that we get not only high significance but the expected signs on coefficients for GDP

and Imports in the cointegrating portion of (3b). These signs are consistent with Figure 1, which suggests WIR

activity tends to rise with GDP, but is depressed by the portion going to Imports. Bounds and cointegration

tests in rows (a) and (b) are all highly significant, and the null of no serial correlation cannot be rejected by any

p-value in row (c).

III. WIR Trade Credits: Registered and Non-Registered Members

According to WIR-Bank statistician Stefan Winkler (2010), WIR client-companies are a large part of the

Swiss total in several industrial sectors, as Table 4 shows. Previously unreleased data are for 2005, the last year

for which nation-wide totals were made available by the WIR-Bank. Notice that the number of Non-Registered

Clients is two or three times that of Registered Clients in all sectors but Hospitality. According to Winkler

(2010), this Non-Registered group includes some very large corporations. WIR-Bank cannot list the names of

these companies due to Swiss bank secrecy laws (Winkler, 2010). Large firms, for their part, cannot become

WIR members: a 1972 by-law stated that only SMEs can be members (Stodder, 2009). Registered firms must

accept WIR for at least 30 percent of the payment of their first 2,000 SFr due on a bill. Non-Registered firms

are free to accept any amount – or none at all (Studer, 1998, p. 33).

9

Table 4: WIR-Client Enterprises, by Sector, 2005

All

All

Portion

(1,000 SFr)

(1,000 SFr)

(SFr)

Turn/Balance =

Industry

Swiss

WIR

WIR/Swiss

Turnover

Tot. Bal.

Av. Bal.

Velocity

RETAIL, of which

62,380

14,275

22.9%

345,757

127,100

8,904

2.720

Registered

5,933

9.5%

223,822

64,958

10,949

3.446

Non-Registered

8,342

13.4%

121,935

62,142

7,449

1.962

SERVICES, of which

164,709

10,380

6.3%

213,515

88,788

8,554

2.405

Registered

3,817

2.3%

112,186

30,745

8,055

3.649

Non-Registered

6,563

4.0%

101,329

58,044

8,844

1.746

HOSPITALITY, of which

28,006

3,438

12.3%

73,021

22,416

6,520

3.257

Registered

2,099

7.5%

61,872

16,156

7,697

3.830

Non-Registered

1,339

4.8%

11,148

6,261

4,676

1.781

CONSTRUCTION, of which

57,268

21,162

37.0%

527,619

210,477

9,946

2.507

Registered

6,992

12.2%

280,169

82,462

11,794

3.398

Non-Registered

14,170

24.7%

247,450

128,015

9,034

1.933

MANUFACTURING, of which

38,421

7,310

19.0%

230,196

101,884

13,938

2.259

Registered

1,820

4.7%

87,418

26,092

14,336

3.350

Non-Registered

5,490

14.3%

142,778

75,792

13,805

1.884

WHOLESALE, of which

21,762

4,138

19.0%

223,631

73,787

17,832

3.031

Registered

1,027

4.7%

80,371

15,462

15,056

5.198

Non-Registered

3,111

14.3%

143,260

58,325

18,748

2.456

TOTALS, of which

372,546

60,703

16.3%

1,613,739

624,452

10,287

2.584

Registered

21,688

5.8%

845,838

235,874

10,876

3.586

Non-Registered

39,015

10.5%

767,901

388,578

9,960

1.976

Source: WIR-Bank Panel Data, 2010. Most of these data, not previously public, may be shared

with interested researchers.

A note on household versus enterprise membership: The total of WIR Client Enterprises shown in Table

4, 60,703, is 81 percent of total for WIR members that year, 74,732, as shown in the annual Rapport de Gestion

(2005). The remaining members are households (Winkler, 2010).

Note in Table 4 that while the Balances of Non-Registered clients are much greater than those of

Registered Clients in all sectors but Retail and Hospitality, the Turnover for both sorts of clients is quite similar

overall, and fairly similar within most sectors. Except for Hospitality, the ratio of high to low Turnover for

Registered and Non-Registered clients is always less than 2. Note however that the Velocity, or

Turnover/Balance, at which Turnover circulates, is always much higher for Registered Clients. As we will see,

this dominance of Velocity for Registered firms (and of Balances for Non-Registered) is fundamental to their

countercyclical activity: Registered firms respond to recessions by increasing the Velocity of WIR Turnover.

10

Non-Registered firms accommodate by allowing their SME customers to settle less in cash and more in WIR –

thus accumulating larger WIR-Balances.

We should note that SMEs typically have less access to formal credit institutions (Terra, 2003), and rely

disproportionately on self-financing (Small Business Administration, 1998) and the trade credits supplied by

larger firms (Petersen and Rajan, 1997; Nilsen, 2002). Following the argument of Studer (1998) on self-

financing trade, WIR-money can be seen as an extension of the trade credits widely used between firms (Greco

2001, p. 68; Stodder, 2009).

Trade credits are traditionally advanced by larger firms to smaller customers and distributors, especially

during recessions (Nilsen, 2002). In the US, for example, trade credits are commonly given on terms of “2% 10,

net 30,” whereby a buyer gets a 2% discount by repaying within 10 days, with full settlement due in 30 days

(Nilsen, 2002). The main use of demand deposit accounts for most businesses, according to Clower and Howitt

(1996, pp. 26-28), is to clear such trade credits. By accepting delayed payment from a smaller firm via a trade

credit, the larger firm thereby accumulates a credit in its accounts receivable – an increase in its trade credit

balance.

In a Philadelphia Fed publication, Mitchel Berlin (2003) notes that there has been surprisingly little

macroeconomic study of trade credits, despite the fact that they are the primary form of countercyclical credit

for SMEs. Petersen and Rajan (1994, 1997) find that only 11 to 17 percent of large-firm assets in the G7

countries are dedicated to accounts payable, but 13 to 29 percent in accounts receivable. Since accounts

receivable normally exceed accounts payable for large firms, this implies an extension of trade credit. This same

inequality shows that receiving trade credits is more important than granting them for smaller firms, in their role

as customers or distributors for larger ones.

Nilsen (2002) finds that use of trade credits is countercyclical for SMEs. SMEs are more likely to be

credit-rationed by banks when money is tight, leaving trade credits as their only form of credit. The importance

of trade credits for the macro-economy was indirectly shown by the Federal Reserve’s decision to buy up

“commercial paper” in the 2007-2009 financial crisis, thereby lowering short-term finance costs for large firms.

11

The link between commercial paper for large firms, and the trade credits they in turn provide to smaller

customers, is noted by Bernanke and Gertler (1995, p. 38, ff. 15). In a recent World Bank publication, Love

notes that if there is severe disruption to the commercial paper market, “there may be nothing left to redistribute

through trade credit.” (2011, p. 34)

In the case of WIR-credits for commercial trade, many types of goods and services are exchanged –

construction, hotel stays, restaurant meals, used vehicles, legal services – with offerings posted online and in

publications like WIR-Plus (2000-13). Prices are quoted in both Swiss Francs (SFr) and WIR, and often a mix

of the two, and with a maximum posted for the percent of payment accepted in WIR. For ease of comparison,

WIR prices are denominated in the same units as SFr. The WIR-Bank keeps account of each customer in terms

of her credits or debits. This is partly to check that large sums of WIR are not being traded for cash; i.e., that

there is a rough balance of credits and debits over time. From the individual’s point of view, an account in WIR

is much like an ordinary checking account with clearing Balances and limits on how large a negative Balance or

“overdraft” can be run. WIR-Bank is a registered Swiss bank, and so also provides ordinary banking services in

SFr.

Non-Registered clients are not subject to the organization’s by-laws and thus not obliged to accept any

minimum share of payment in WIR (Studer, 1989, p. 33) – as Registered clients must do. The Non-Registered

are thus free to extend the privilege of WIR-settlement to most favored customers, or only when it is most

needed – as during an economic downturn. This helps explain the countercyclical variability in Non-Registered

accounts, a role similar to that of trade credits. If the relationship of Non-Registered to Registered firms is

predominantly that of suppliers to customers/distributors, then we would expect to see countercyclical responses

showing as increased Balances for the former and by increased Velocities for the latter. We will show this

pattern in the econometric tests on Registered and Unregistered firms.

Yet we must also note here two crucial differences between ordinary trade credits and WIR-credits.

First, unlike an ordinary trade credit payable in Swiss Francs, a payment in WIR is itself final payment. As

long as the WIR-Bank exists, a firm receiving WIR for it products will never see the check “bounce.” Second,

12

like any true currency, WIR-credits support multilateral, not just bilateral exchange. That is, a WIR-creditor’s

value is ensured, not by her debtor’s ultimate willingness to settle in cash, but by the current willingness of

thousands of firms and households to accept WIR as final payment. To repeat Studer’s formulation (1998, p.

32), “every franc of WIR credit automatically and immediately becomes a franc of WIR payment medium.”

Since every WIR-credit is matched by an equal and opposite debit, the system as a whole must net to

zero. Individual traders can have positive or negative Balances or “overdrafts.” The latter is, in effect, a loan

from the WIR-Bank. Short-term overdrafts are interest-free, with an overdraft limit “individually established”

(Studer, 1998, p. 31) by credit history. As long as these limits are maintained, the WIR-Bank can be quite

relaxed about variations in its total bank Balances. The system is also highly flexible: while the net value of

WIR credits and debits must be zero, their absolute total is determined by current economic activity alone –

there is no monetary base as such. This balanced flexibility of an “automatic plus-minus balance of the system

as a whole” (Studer 1998, p. 31) is shown in teaching software for a system similar to WIR, as devised by

Linton and Harris-Braun (2007), available at www.openmoney.org/letsplay/index.html. In this program,

balances increase in the alternative currency as traders gain confidence in the system and are able to liquidate

unsold inventories.

A second difference with trade credits is that WIR-exchange is totally centralized, combining the

functions of the commercial bank system and a central bank. The WIR-Bank thus has much more detailed

knowledge of credit conditions in its own currency than any ordinary commercial or central bank. Of course it

can still make mistakes, extending too many overdrafts or direct loans. Such credit "inflation" has occurred in

WIR’s history (Stutz, 1984; Defila, 1994; Studer, 1998), but now appears contained by sensible overdraft limits.

The WIR was inspired by the ideas of an early 20th-century German-Argentine economist, Silvio Gesell

(Defila 1994, Studer 1998)

2

. Gesell is given a chapter in Keynes’ General Theory (1936; Chapter 23, Part VI),

2

Gesell was familiar with trade credits from his international trade experience. His use of the term demurrage is borrowed

from international shipping, where it denotes a reduction in payment to compensate for an unscheduled delay in delivery. Similarly,

Gesell applies demurrage to the holding of currency balances, with the aim of increasing velocity.

A form of bank-mediated trade credit common in international trade is the banker’s acceptance, which allows the exporter to

be paid upon embarkation, while the importer does not have to pay until taking possession of the goods. Credits from the WIR-bank

13

whom he sees as an “unduly neglected prophet,” anticipating some of his own ideas on why interest rates may

exceed the marginal efficiency of capital. Keynes notes (1936, p. 355) that “Professor Irving Fisher, alone

amongst academic economists, has recognised [Gesell’s] significance,” and predicts that “the future will learn

more from the spirit of Gesell than from that of Marx.” Although the intellectual linkage between Keynesian

and Gesellian ideas has received substantial attention (Klein, 1980; Darity, 1995) Gesellian institutions like the

WIR-Bank have not.

IV. Theory: Money in the Production Function

Stodder (2009) formalizes the interaction of WIR-money and national currency via “money in the

production function” (MIPF). This is analogous to “money in the utility function” (MIUF), and similarly

derived by the implicit function theorem. Both MIPF and MIUF are justified by the transactions-cost-saving

role money plays, moving the economy to its efficiency frontier. There is a substantial literature on this idea

(Patinkin, 1956; Fischer, 1974, 1979; Finnerty, 1980; Rösl, 2006).

We formalize the basic result by showing a profit-maximizing firm as minimizing both its direct and

transactional costs subject to the constraint of producing quantity,

Q

, exogenously determined by the market:

Min: cpKp + csKs + rpMp + rsMs (1)

s.t.:

Q

=

p

Q

+

s

Q

≤ f(Kp, Mp, Ks, Ms) = fp[(

sp KK ,

), Mp] + fs[(

sp KK ,

), Ms].

Here the primary national and secondary social currency, Mp and Ms, show interest rates/opportunity costs of rp

and rs. They are used to pay the market costs, cp and cs, of the required capital inputs, Kp and Ks, respectively.

These inputs are assumed divisible and perfect substitutes. Subscripts account only for means of purchase, since

many purchases are for a mix of WIR and SFr (Studer, 1989, p. 33). The production/transaction functions

p

Q

=

fp[(

sp KK ,

), Mp] and

s

Q

= fs[(

sp KK ,

), Ms], are assumed concave and differentiable, with the bars indicating

that Output quanties

p

Q

and

s

Q

are set exogenously. The Marginal Rates of Substitution (MRS) derived from

(1) show that inventories of money and physical inputs can be substitutes.

can be interpreted as extending the payment terms on bankers’ acceptances from short to medium-term, and from bilateral to

multilateral.

14

It is assumed that rp > rs and cp ≤ cs. The first inequality arises because primary currency is more useful

than secondary, and so must have a higher opportunity cost. This is recognized by Studer (1998, p. 31), who

states it as a basic fact about WIR commerce: “Since the WIR Bank operates in competition with conventional

credit banks and a WIR loan is less universally useful than a cash loan, the cost of WIR credit must in any case

be kept lower than normal interest rates.” The second inequality arises from the same fact: since WIR money is

less useful, items for sale are often posted at WIR prices higher than their equivalent price in SFr. (Stodder,

2009).

Lemma 1: For a cost minimizing firm, the marginal productivity of Ks is at least as great as that for Kp,

but that of Ms is less than Mp.

Proof: Using the above inequalities, first order conditions of (1) yield

(cs/cp) = (∂f/∂Ks)/(∂f/∂Kp) ≥ 1 > (rs /rp) = (∂f/∂Ms)/(∂f/∂Mp). (2)

Secondary currencies are ‘residual,’ used when the primary currency is unusually scarce. The next

Lemma shows that if Registered (R) clients face more restricted credit conditions than larger Non-Registered

(NR) clients (that is, a higher interest rate on primary money, rp), then larger holdings of Ms/Mp Balances for

Registered than Non-Registered clients will result:

Lemma 2: If a Registered firm (R) is more credit constrained in primary currency than a Non-Registered

firm (NR),

, yet their access to secondary currency is equivalent,

then R’s holdings of the

secondary currency must be relatively larger:

.

Proof: The above assumptions give rsR/rpR < rsNR/rpNR. By Lemma 1, each ratio is equal to the ratio of

marginal products of secondary to primary currency, (∂f/∂Ms)/(∂f/∂Mp) for Registered and Non-Registered

firms, respectively: (∂f/∂MsR)/(∂f/∂MpR) < (∂f/∂MsNR)/(∂f/∂MpNR) Since the production/ transaction function f(

) is assumed concave and the same for each firm, R must therefore hold a larger ratio of secondary to primary

currency than NR.

15

Table 4 showed that Registered firms hold larger average WIR Balances than Non-Registered firms. If

the average SFr holdings for Registered firms are also smaller than for Non-Registered firms (Winkler, 2010),

then a fortiori, they must have larger relative Balances of WIR to SFr., as in Lemma 2.

Smaller Registered clients may be quite limited in their access to credit in the primary currency (Winter-

Ebmer and Zweimüller, 1999). In a recession, SMEs may lose such credit altogether (Wan et. al., 2011). As

with the greater use of supplier-provided trade credits by SMEs during a recession (Nilsen, 2002), so a greater

portion of WIR currency is often accepted by larger Non-Registered firms in a recession, thus helping their

Registered SME customers to conserve cash.

Note that Table 4 shows Non-Registered firms with WIR Turnover levels comparable to those of the

smaller Registered firms. This suggests that large Non-Registered firms limited WIR activity almost

exclusively to their smaller Registered customers, Registered firms playing a reciprocal role. This rough parity

of Turnover can be shown to hold for the 15 years of our sample.

Given this pattern, let us assume for modeling purposes, not too unrealistically, that secondary currency

trade between Registered and Non-Registered firms is “mirror-imaged” in the sense that Turnover is equal for

both types, but Balances for each type show opposite cyclical effects. Countercyclical activity is structured by

such reciprocity, as we show in Lemma 3:

Lemma 3: Consider the elasticity of secondary currency expenditures with respect to Output. Let

(i) Turnover Elasticity for both Registered (R) and Non-Registered (NR) firms be of counter-cyclial

(negative) sign, and

(ii) NR firms allow R firms to settle a greater proportion of bills outstanding in secondary currency in a

recession, but wait to spend most of this currency until the recession is over.

It follows that:

(1) For an R firm, its countercyclical Turnover is dominated by a countercyclical Velocity. That is,

although both are negative, the Velocity Elasticity of Output is greater in absolute value than the Turnover

Elasticity of Output.

16

(2) For an NR firm, its countercyclical Turnover is dominated by a countercyclical Balance. That is,

although both are negative, the Balance Elasticity of its Output is greater in absolute value than the Turnover

Elasticity of its Output.

Proof: By (ii) we have countercyclical NR Balances, and pro-cyclical R Balances. Thus we can write

; i.e., the Balance (B) Elasticity of Output (Q) is countercyclical for NR, but pro-cyclical for R

firms. Since (i) Turnover is countercyclical, this gives our first result:

, (3.1)

(-) (+)

or the countercyclical Turnover of an R firm is dominated by its countercyclical Velocity. Again by (ii), R

Velocity is countercyclical, and that of NR is pro-cyclical:

Again, countercyclical Turnover

(i) gives our second result:

, (3.2)

(+) + (-)

or the countercyclical Turnover of an NR firm is dominated by its countercyclical Balance.

V. Panel Econometric Tests

V.1 A Bootstrap Simulation to Test Cointegration

Instead of regressing overall WIR activity against Swiss GDP, as in Table 3 above, more disaggregated

data allow us to regress WIR activity against changes in GDP by sector. Our six WIR sectors were shown in

Table 4; we now regress them against GDP Value-Added by sector. Our sectoral time series is short, 15 years

only, so we are not so concerned with the “long-term” relationship between imports and WIR, or with the EC

portion of the VEC. As long as the EC equation is cointegrated, we can concentrate on the coefficients of the

lagged, first-differenced terms – the VAR part of the model, where any countercyclical effects would show.

Fisher tests on moving average terms overwhelmingly fail to reject the null hypothesis of individual unit

roots. The 5 percent rejection level was achieved for the null of a common unit root across the panel – by the

Levin, Lin, and Chu, or LLC test – for two series: the moving average log of real WIR Velocities and Turnover

for Non-Registered firms. The null of individual unit roots across the panel, however – using the Im-Pesaran-

17

Shin (IPS), Augmented Dickey Fuller, and Phillips-Perron tests – could not be rejected at 5 percent. These

latter two tests are versions of the Fisher test. Maddala and Wu (1996) report that in the presence of correlation

between sectors, as we have here, Fisher tests have greater power to reject the null than either the LLC common

or the IPS individual test. We undertook 14 Fisher-type tests – 2 each for the moving averages of Turnover,

Velocity, and Balances on both Registered and Non-Registered firms, plus 2 for the moving average of Output,

or 2 x 3 x 2 + 2 = 14. Only one test on Registered Balances and one on Non-Registered Turnover could reject

the null of a unit root at 10 percent; no others could reject even at 15 percent. With strong evidence of unit

roots, cointegration tests are legitimate.

Panel cointegration, however, is problematic with our small data set, with only 15 years and 6 sectors,

since there will generally have poor size characteristics when correlation between panel sectors exists (Banerjee

and Carrion-i-Silvestre, 2006) – as can easily be shown to be the case here. A common resort is to extend the

time dimension through bootstrapping. We have done this here with a Residual-Based Stationary Bootstrap

(RSB) method for cointegration testing by Di Iorio and Fachin (2011).

We begin by residual-based cointegration through successive residuals, εt and εt-1, taken from a

cointegrating equation, , where and have unit roots. We then form an AR(1) relation

on the residuals, The estimated residuals,

, are then “re-shuffled” in chained blocks of

random length, at random locations. Replacing these estimated residuals,

, with their reshuffled pseudo-

residuals,

, there is a low probability that

=

. We use the

term to replace our estimated AR residuals,

with a new set of AR-pseudo-residuals,

, based on the null hypothesis of no-cointegration;

i.e., , where

for the first element of the series. The Data Generating Process (DGP) in

bootstrapping should mimic the null hypothesis (Maddala and Kim, 1998, p. 317; Van Giersbergen and Kiviet,

1994). The panel cointegration tests used here are appropriate by this criterion, since they are based on the null

of no cointegration.

Armed with these AR-pseudo-residuals, we simulate new values of our dependent variable,

and where and

are the estimates from our original cointegrating equation. This process is

18

repeated B-1 times, until the time dimension of our pseudo-data is not T but BT. (In the simulations below, B =

13.) Note that the dependent variable, – not the original independent variable, – is simulated by this

bootstrap. If we want to reverse the role of and in cointegration, we must perform a new bootstrap.

Using this simulation method, a full battery of Pedroni cointegration tests was implemented. A lag

length of 1 was used, consistent with cointegrating equations to follow. An example of a Pedroni test is seen in

Table 5. Here LrTURNa_NR(i,t) = the log of real WIR Turnover averaged over 2 years for Non-Registered

Clients for i = 1, 2, .., 6 industrial sectors, and LrVAa(i,t), the log of averaged Real Output (Value-Added, in

Swiss Francs), again for each sector. The original t = 1, 2, .., 15 periods were expanded by 12 simulated

sequences of 15, yielding a total of 13x15 =195 periods, with 6 sectors for 15x13x6 =1170 observations. (A

few dozen observations are missing because there were no 1994 observations on Balances or Velocity. These

lacunae are magnified by the simulation’s re-shuffling.)

Note that the null hypothesis of No Cointegration is more decisively rejected (i.e., lower p-Values) in

simulations where the alternative hypothesis was for “Group” cointegration; i.e., AR residuals for each sector i,

formed by separate coefficient values < 1. This is as compared to an alternative hypothesis of “Panel”

cointegration, with a common value across all sectors. Pedroni (2001) refers to these as the “between”

and “within” dimension, respectively.

Table 5. Pedroni Residual Cointegration Tests: LrTURNa_NR(i,t) = α + β*LrVAa + ε(i,t);

Bootstrapped data, DOLS Regression for Non-Registered WIR Clients.

(Null Hypothesis: No cointegration.)

Alternative hypothesis:

common AR coefficients, (within-dimension)

Weighted

Statistic

P-Value

Panel v-Statistic

32.2372

0.00E+00

Panel rho-Statistic

-59.6962

0.00E+00

Panel PP-Statistic

-24.4248

4.66E-132

Panel ADF-Statistic

-4.6886

1.38E-06

Alternative hypothesis:

individual AR coefficients, (between-dimension)

Statistic

P-Value

Group rho-Statistic

-57.6423

0.00E+00

Group PP-Statistic

-29.3090

3.98E-189

Group ADF-Statistic

-4.6491

1.67E-06

Data: 195 periods = 15x13, 6 Cross-sections => 15x13x6 =1170 Observations, some omitted by re-shuffling. In

Pedroni tests, no deterministic trend, Lag length = 1. Newey-West bandwidth selection and Bartlett kernel.

19

Despite somewhat lower P-values for this Group or “between” case – a consistent finding – we focus

here on the Panel or “within” case for the following reasons. First, WIR clients trade with other sectors, not just

their own. Second and more importantly, our interest is not just in the intra-sectoral circulation of WIR, but its

broad countercyclical effectiveness. It would be natural for firms to respond first to business conditions within

their own sector, but myopic for them to look only there.

Following the example of Table 5, cointegration tests were carried out in 24 simulations: 2 types of

clients, Registered and Non-Registered, 2 cointegration forms, FMOLS and DOLS, 3 possible pairings of 3

variables, and 2 options as to which is dependent and which the independent variable in each pairing. A

summary of these 2 x 2 x 3 x 2 = 24 possible relations is shown in Table 6.

For the cointegration tests carried out for Table 6 we used 1 lag, consistent with the VEC regression

models to follow. Also for consistency with these VEC regressions, with 3 lags in their VAR portion, we use 4

lags for the Granger tests of Table 6. This is based on the Toda-Yamamoto (1995) criteria of the number of

VAT lags one for each possible cointegrating relationship.

In Table 6, the highest P-value for the Weighted Panel statistics is used for the sake of conservatism. For

example, the ADF Statistic in Table 5 shows a value of 1.38E-06. This is the first DOLS p-value shown in

section 4, column 2 of Table 6.

Table 6 is divided into white and grey cells to emphasize the pattern of cointegration within the white,

but not the grey cells. Note that cointegration P-values in grey cells are all above the 5 percent level. P-values

significant at 5 percent are written here in scientific notation.

Sections 1-3 show the cointegration and Granger P-values for R firms, while Sections 4-6 are for NR

firms. For Section 1, recall that our earlier time-series estimates in Table 3 showed WIR-Turnover as both

cointegrated and countercyclical with GDP. Table 6 shows sectoral Value-added; i.e., the sectoral contribution

to GDP. Thus we expect these entries to show significant cointegration and Granger precedence – as they do.

20

Table 6. Highest P-Value of 4 Pedroni Panel Cointegration Tests; P-values in Bold Italics; Null of No

Cointegration. P-values on Granger Precedence; (Reverse Precedence in Parentheses); Null of No Granger

Precedence.

Section

#

Dependent

Variable in

Dependent

Variable in

Simulation

FMOLS

DOLS

FMOLS

DOLS

Simulation

(1)

Registered:

LrVAa =>

LrTURNa

2.33E-02

3.15E-02

1.79E-10

0.1882

Registered:

LrVAa <=

LrTURNa

6.79E-02

0.1107

0.7589

0.2793

(0.5472)

(0.4092)

(2.39E-06)

(1.64E-04)

(2)

Registered:

LrVAa =>

LrVELa

7.19E-05

2.47E-14

0.00E+00

3.52E-10

Registered:

LrVAa <=

LrVELa

3.87E-06

2.68E-09

3.97E-02

0.6226

(0.2210)

(0.2760)

(0.9637)

(7.98E-02)

(3)

Registered:

LrVAa =>

LrBALa

0.6180

0.9644

8.17E-09

2.17E-11

Registered:

LrVAa <=

LrBALa

4.80E-03

0.1063

4.72E-02

5.00E-02

(4.05E-03)

(2.01E-02)

(0.5758)

(6.97E-06)

(4)

Non-Registered:

LrVAa =>

LrTURNa

2.05E-02

1.38E-06

3.47E-20

1.87E-11

Non-Registered:

LrVAa <=

LrTURNa

0.00E+00

9.33E-265

0.1582

0.4011

(1.75E-36)

(9.73E-35)

(4.68E-06)

(6.94E-06)

(5)

Non-Registered:

LrVAa =>

LrVELa

0.4782

0.9871

3.89E-08

3.34E-11

Non-Registered:

LrVAa <=

LrVELa

5.49E-09

9.14E-07

0.1022

4.73E-03

(0.3950)

(0.0932)

(1.72E-03)

(6.44E-02)

(6)

Non-Registered:

LrVAa =>

LrBALa

3.07E-03

2.96E-06

8.65E-07

1.62E-11

Non-Registered:

LrVAa <=

LrBALa

2.24E-11

0.00E+00

2.00E-04

1.25E-05

(0.1432)

(3.59E-13)

(7.00E-07)

(1.86E-05)

Data: Bootstrap simulation of panel data: 15 periods simulated more 12 times, 6 Cross-sections

=>15x13x6 = 1170 observations; some omitted by simulation re-shuffling. P-values < 5 percent are

given in scientific notation. Lags of 1 used for Cointegration tests, and lags of 4 for the Toda-Yamamoto

Granger tests, as explained in text. Greyed-out area show Cointegration test p-values > 5 percent.

It is somewhat surprising that cointegration and Granger relations are seen also in the opposite direction,

in the right-side columns of Section 1, where Value-added is now a dependent variable. Note here that Granger

precedence appears to run from Value-added to WIR-Turnover, but not in the opposite direction. This is

unsurprising, given the size of WIR Turnover relative to Swiss GDP.

Compere this pattern for Registered in Section 1 with Turnover for Non-Registered firms in Section 4.

The first two column entries in both Sections show the expected cointegration relations, consistent with our

time series estimates in Table 3. In the opposite direction, in the right-hand columns, there is again

21

cointegration and Granger precedence. But as with the Registered firms, note that WIR Turnover does not

appear to Granger precede Value-added.

For Sections 2 and 6 in the left-hand columns, recall that Lemma 3 predicted countercyclical Velocities

for R firms, and countercyclical Balances for NR firms. These relationships may be reflected in the strong

evidence for cointegration. When we move to the right-hand columns, we find that R firms show Granger

precedence from Value-added to WIR-Velocity, but evidence for the reverse precedence is mixed. NR firms,

however, now show cointegration and Granger precedence in both directions between WIR-Balances and

Value-added. This probably reflects the larger scale of the NR firms, restricted from direct WIR membership

(but not client relations) by WIR-Bank bylaws.

Finally, for the greyed-out Sections 3 and 5, recall that Lemma 3 predicted that Velocity for NR firms

and Balances for R firms will be pro-, not countercyclical. Thus we expected (although it is not strictly implied)

that these variables would not show cointegration with Value-added – and instead show a looser ‘residual’

relationship. Very high p-values are indeed shown for all cointegration entries in these greyed-out areas. In the

right-side columns of the same Sections 3 and 5, however, while there is again strong evidence of cointegration,

the direction of Granger precedence is ambiguous. The only unambiguous evidence of direct macro-economic

impact from WIR activity appears in the last two columns of Section 6: the WIR Balances of Non-Registered

firms appear to Granger precede changes in Value-added.

In summary, the leading countercyclical role sketched for NR Balances and R Velocities is strengthened

by these cointegration and Granger results. In the case of the larger NR firms, there is also some evidence for

‘reverse causality’ – WIR Balances for NR firms may be large enough to have a direct macro-economic impact

on the larger Swiss economy. This is a theme to which we will return.

22

V.2 Vector Error Correction (VEC) Estimates

V.2.1 VECs for Registered Firms

In Table 7, for Registered WIR-Clients, the cointegrating equations shown here are from the actual data,

not the bootstrap results shown in Table 6. From the Pedroni cointegration and Granger tests from Table 6,

however, reproduced here for convenience, cointegration modeling is clearly valid.

Results in Table 7 are encouraging: the coefficients on the first-differenced Value-Added terms in

columns (1a) and (1b) are of the right countercyclical sign, though not highly significant. Note that the

significant countercyclical effects in columns (2a) and (2b), are of the opposite sign, as they must be for

stability, but only on the second lag.

In columns (1a) and (1b) of Table 7, the Wooldridge (2002) null hypothesis of no first-order serial-

correlation cannot be rejected; thus serial correlation is not a likely problem. In columns (2a) and (2b),

however, this null of no first-order auto-regression is strongly rejected. This is not quite as bad as it seems,

however. In Table 7 and the panel regressions that follow, we use White (1980) period estimators, robust to

within-cross-section serial correlation (Arellano, 1987). This means our coefficient estimates are unbiased but

not efficient; i.e., their standard errors are not as small as possible. So despite our serial correlation result,

significance levels are conservative and we can be fairly confident about signs of these coefficients. It is

therefore credible that the coefficients – here in bold – on the second-lagged, first-differenced Turnover in

columns (2a) and (2b) show positive and highly significant effects on Value-added. From Table 6, however, we

know that although there is cointegration, there is no evidence of Granger precedence in this direction – from

Turnover to Value-added.

23

Table 7: REGISTERED WIR Clients: 2 Year Moving Averages of Log of Real WIR Turnover

(LrTURNa), regressed on Log of Real Value-Added (LrVAa), by Sector

t-statistics in [ ]; ***: p-val ≤ 0.01, ** : p-val ≤ 0.05, *: p-val ≤ 0.10

Method: Vector Error Correction Model, Panel Data, Fixed Effects

White Period Covariance (no degrees of freedom correction)

Sample (adjusted): 1999-2008

Periods: 10, Cross-sections: 6

Sample (adjusted): 1999-2008

Periods: 10, Cross-sections: 6

COINTEGRATING

EQUATION

(METHOD)

(1a)

(1b)

(2a)

(2b)

Depend. Variable:

Depend. Variable:

Depend. Variable:

Depend. Variable:

LrTURNa(-1)

(FMOLS)

LrTURNa(-1)

(DOLS)

LrVAa(-1)

(FMOLS)

LrVAa(-1)

(DOLS)

Constant

12.27

9.670

12.24

8.453

LrVAa(-1)

0.0820

0.3877

LrTURNa(-1)

[0.33]

[1.03]

-0.1641

0.1781

[-0.53]

[0.28]

Independent Variables:

Depend. Variable:

Depend. Variable:

Depend. Variable:

Depend. Variable:

D(LrTURNa)

D(LrTURNa)

D(LrVAa)

D(LrVAa)

CointegEqRES

-0.1241

-0.0452

-0.0542

0.0609

[-1.57]

[-1.22]

[-0.72]

[3.70]***

D(Dependent Var. (-1))

0.6747

0.5755

1.1834

1.0393

[6.60]***

[10.01]***

[5.97]***

[5.87]***

D(Dependent Var. (-2))

-0.5485

-0.5704

-0.8079

-0.7856

[-18.91]***

[-29.92]***

[-3.87]***

[-3.83]***

D(Dependent Var. (-3))

0.3374

0.2661

0.3716

0.2327

[3.37]***

[3.97]***

[2.57]**

[1.98]*

D(LrVAa(-1))

-0.7861

-0.5249

D(LrTURNa(-1))

[-1.81]*

[-1.56]

-0.0280

-0.0251

[-0.70]

[-0.97]

D(LrVAa(-2))

0.2867

0.2233

D(LrTURNa(-2))

[0.82]

[0.55]

0.0676

0.0568

[3.21]***

[3.12]**

D(LrVAa(-3))

0.3095

0.4802

D(LrTURNa(-3))

[1.01]

[1.534]

0.0071

0.0103

[0.230]

[0.51]

Constant

-0.1771

-0.1023

0.0060

0.0254

[-1.79]*

[-1.54]

[2.20]**

[4.34]***

R-squared

0.687

0.669

0.706

0.717

Adjusted R-squared

0.607

0.585

0.631

0.645

Log likelihood

112.6

111.0

189.5

190.7

F-statistic

8.579

7.930

9.414

9.931

Akaike info criterion

-3.321

-3.268

-5.884

-5.922

Schwarz criterion

-2.867

-2.814

-5.430

-5.468

a) Pedroni Tests:

2.33E-02

3.15E-02

1.79E-10

0.1882

b) Wooldridge AR1:

0.4765

0.3740

0.0000

0.0000

c) Granger Precedence:

6.79E-02 (0.5472)

9.33E-265 (9.73E-35)

0.7589 (2.39E-06)

0.2793 (1.64E-04)

Notes: P-values in a)-c) are based on null hypotheses of: a) No panel cointegration, from estimates of Tables 6; b) No

first-order serial-correlation (Wooldridge AR test); and c) No Granger Precedence. For c), the first p-value tests if the

independent variable does not Granger precede the dependent variable. (The p-value in parentheses is for the reverse

precedence.)

24

Table 8: REGISTERED WIR Clients: 2 Year Moving Averages of Log of Real WIR Velocity (LrWirVelAv2),

regressed on Log of Real Value-Added (LrVaAv2), by Sector

t-statistics in [ ]; ***: p-val ≤ 0.01, ** : p-val ≤ 0.05, *: p-val ≤ 0.10

Notes: P-values in a)-c) are based on null hypotheses of: a) No panel cointegration, from estimates of Tables 6; b) No

first-order serial-correlation (Wooldridge AR test); and c) No Granger Precedence. For c), the first p-value tests if the

independent variable does not Granger precede the dependent variable. (The p-value in parentheses is for the reverse

precedence.)

Turning to Table 8, we regress Velocity against Value-Added, and we can compare the countercyclical

coefficients to Table 7. As predicted by Lemma 3, the countercyclical coefficients on Velocity in Table 8 are

Method: Vector Error Correction Model, Panel Data, Fixed Effects

White Period Covariance (no degrees of freedom correction)

Sample (adjusted): 1999-2007

Periods: 9, Cross-sections: 6

Sample (adjusted): 1999-2008

Periods: 10, Cross-sections: 6

COINTEGRATING

EQUATION

(METHOD)

(1a)

(1b)

(2a)

(2b)

Depend. Variable:

Depend. Variable:

Depend. Variable:

Depend. Variable:

LrVelAv2(-1)

(FMOLS)

LrVelAv2(-1)

(DOLS)

LrVaAv2(-1)

(FMOLS)

LrVaAv2(-1)

(DOLS)

Constant

10.86

14.10

12.24

11.22

LrVaAv2(-1)

-0.3963

-1.137

LrVelAv2(-1)

[-3.71]***

[-2.52]**

-0.1641

-0.6328

[-0.53]

[-6.37]***

Independent Variables:

Depend. Variable:

Depend. Variable:

Depend. Variable:

Depend. Variable:

D(LrWirVelAv2)

D(LrWirVelAv2)

D(LrVaAv2)

D(LrVaAv2)

CointegEqRES

-0.697

-0.220

1.1610

-0.014

[-9.63]***

[-2.82]***

[8.26]***

[-0.37]

D(Dependent Var. (-1))

0.638

0.358

-0.7004

1.1158

[5.13]***

[2.56]**

[-12.11]***

[9.79]***

D(Dependent Var. (-2))

-0.152

-0.396

0.3170

-0.7007

[-1.41]

[-6.45]***

[6.75]***

[-9.87]***

D(Dependent Var. (-3))

0.334

0.217

1.1610

0.2881

[3.58]***

[2.91]***

[8.26]***

[7.18]***

D(LrVaAv2(-1))

-0.856

-1.871

D(LrWirVelAv2(-1))

[-4.74]***

[-3.63]***

0.0146

0.0030

[0.51]

[0.15]

D(LrVaAv2(-2))

0.821

1.472

D(LrVelAv2(-2))

[1.55]

[1.52]

0.0182

0.0094

[0.76]

[0.47]

D(LrVaAv2(-3))

-0.683

[-2.17]**

-0.175

[-0.22]

0.0747

0.0681

D(LrVelAv2(-3))

[2.64]**

[2.74]***

Constant

-0.106

-0.234

0.006

0.007

[-12.42]***

[-3.09]***

[2.13]**

[2.29]**

R-squared

0.728

0.471

0.772

0.765

Adjusted R-squared

0.649

0.316

0.714

0.704

Log likelihood

90.78

72.79

197.1

196.2

F-statistic

9.161

3.042

13.26

12.72

Akaike info criterion

-2.880

-2.214

-6.138

-6.105

Schwarz criterion

-2.402

-1.735

-5.684

-5.652

a) Pedroni Tests:

2.47E-14

7.19E-05

0.981

0.858

b) Wooldridge AR1:

0.000

0.024

0.000

0.000

c) Granger Precedence

1.89E-4 (2.23E-10)

0.552 (3.06E-03)

25

greater than those on Turnover in Table 7. We will shortly undertake formal tests of these differences. As in

Table 7, we can accept the sign and significance of these variables, despite the rejection of no serial correlation

by the Wooldridge tests. Also as in Table 7 when WIR activity is the independent variable in columns (2a) and

(2b), the significant coefficients are at later lags, and evidence for Granger precedence is ambiguous.

V.2.2 VECs for Non-Registered Firms

Let us turn now to Table 9, which shows the relation between sectoral WIR Turnover and sectoral

Value-Added for Non-Registered WIR clients. In the left-hand columns, we see that the coefficient on the first

lagged Value-Added term – in bold – is of countercyclical sign and highly significant. In the right-hand

columns significant coefficients on Turnover are shown only on the second or third lag – similar to the pattern

in Table 7. Note also that row c) shows Granger precedence, but not in both directions. Granger precedence

runs from Value-added to WIR-Turnover for Non-Registered firms, but not the opposite way.

Turning to Table 10, we find that the first and second lagged Value-Added terms – in bold – are not

significant in (1a) and (1b), whereas the WIR-Balance coefficients in (2a) and (2b) show significance only at a

later lag – repeating the pattern shown in Tables 7-9. Unlike these previous results, however, now there is now

strong evidence of Granger precedence in both directions. The implication is that the WIR-activity of Non-

Registered firms, probably because of their larger scale, has a direct countercyclical impact on GDP, and is not

just a reflection of broader macro-economic trends.

26

Table 9: NON-REGISTERED WIR Clients: 2 Year Moving Averages of the Log of Real WIR Turnover

(LrTurnAv2), regressed on Log of Real Value-Added (LrVaAv2), by Industrial Sector

t-statistics in [ ]; ***: p-val ≤ 0.01, ** : p-val ≤ 0.05, *: p-val ≤ 0.10

Method: Vector Error Correction Model, Panel Data, Fixed Effects

White Period Covariance (no degrees of freedom correction)

Sample (adjusted): 1999-2008

Periods: 10, Cross-sections: 6

Sample (adjusted): 1999-2008

Periods: 10, Cross-sections: 6

COINTEGRATING

EQUATION

(METHOD)

(1a)

(1b)

(2a)

(2b)

Depend. Variable:

Depend. Variable:

Depend. Variable:

Depend. Variable:

LrTurnAv2(-1)

(FMOLS)

LrTurnAv2(-1)

(DOLS)

LrVaAv2(-1)

(FMOLS)

LrVaAv2(-1)

(DOLS)

Constant

18.32

16.60

13.59

14.67

LrVaAv2(-1)

-0.5240

-0.3106

LrTurnAv2(-1)

[-1.71]*

[-1.33]

-0.2725

-0.3613

[-4.36]***

[-7.08]***

Independent Variables:

Depend. Variable:

Depend. Variable:

Depend. Variable:

Depend. Variable:

D(LrTurnAv2)

D(LrTurnAv2)

D(LrVaAv2)

D(LrVaAv2)

CointegEqRES

-0.1143

-0.0479

-0.1651

-0.1165

[-4.92]***

[-3.13]***

[-2.58]***

[-2.19]**

D(Dependent Var. (-1))

0.2232

0.2143

0.0263

0.9910

[1.73]*

[1.54]

[1.08]

[8.11]***

D(Dependent Var. (-2))

-0.0453

-0.0950

0.1292

-0.6131

[-0.47]

[-1.05]

[3.39]***

[-4.68]***

D(Dependent Var. (-3))

0.2175

0.1935

0.0357

0.1965

[4.04]***

[3.07]***

[7.41]***

[2.12]**

D(LrVaAv2(-1))

-0.6702

-0.5192

D(LrTurnAv2(-1))

[-4.34]***

[-3.58]***

0.0263

0.0317

[1.08]

[1.52]

D(LrVaAv2(-2))

0.4359

0.4664

D(LrTurnAv2(-2))

[2.06]**

[1.78]*

0.1292

0.0757

[3.39]***

[1.51]

D(LrVaAv2(-3))

-0.0577

-0.0103

D(LrTurnAv2(-3))

[-0.256]

[-0.04]

0.0357

0.0303

[7.41]***

[2.82]***

Constant

-0.1778

-0.1138

-0.0125

-0.0112

[-5.59]***

[-3.83]***

[-1.20]

[-1.12]

R-squared

0.518

0.477

0.7301

0.7136

Adjusted R-squared

0.395

0.343

0.6612

0.6404

Log likelihood

127.2

124.7

192.1

190.3

F-statistic

4.207

3.569

10.59

9.759

Akaike info criterion

-3.806

-3.725

-5.969

-5.909

Schwarz criterion

-3.353

-3.271

-5.515

-5.456

a) Pedroni Tests:

4.66E-132

3.78E-02

3.47E-20

1.87E-11

b) Wooldridge AR1:

0.0006

0.0010

0.0007

0.0002

c) Granger Precedence

0.00E+00 (1.75E-36)

9.33E-265 (9.73E-35)

0.1582 (4.68E-06)

0.4011 (6.94E-06)

Notes: P-values in a)-c) are based on null hypotheses of: a) No panel cointegration, from estimates of Tables 6; b) No

first-order serial-correlation (Wooldridge AR test); and c) No Granger Precedence. For c), the first p-value tests if the

independent variable does not Granger precede the dependent variable. (The p-value in parentheses is for the reverse

precedence.)

27

Table 10: NON-REGISTERED WIR Clients: 2 Year Moving Averages of the Log of Real WIR Balance

(LrBALav2), regressed on Log of Real Value-Added (LrVAav2), by Industrial Sector

t-statistics in [ ]; ***: p-val ≤ 0.01, ** : p-val ≤ 0.05, *: p-val ≤ 0.10

Method: Vector Error Correction Model, Panel Data, Fixed Effects

White Period Covariance (no degrees of freedom correction)

Sample (adjusted): 1999-2007

Periods: 9, Cross-sections: 6

Sample (adjusted): 1999-2007

Periods: 9, Cross-sections: 6

COINTEGRATING

EQUATION

(METHOD)

(1a)

(1b)

(2a)

(2b)

Depend. Variable:

Depend. Variable:

Depend. Variable:

Depend. Variable:

LrBALAv2(-1)

(FMOLS)

LrBALav2(-1)

(DOLS)

LrVaAv2(-1)

(FMOLS)

LrVaAv2(-1)

(DOLS)

Constant

13.24

16.14

12.71

14.32

LrVaAv2(-1)

-0.1892

-0.4242

LrBALAv2(-1)

[-0.75]

[-1.74]*

-0.2141

-0.3534

[-3.08]***

[-6.21]***

Independent Variables:

Depend. Variable:

Depend. Variable:

Depend. Variable:

Depend. Variable:

D(LrBALa)

D(LrBALa)

D(LrVAa)

D(LrVAa)

CointegEqRES

-0.4731

-0.2535

0.0093

-0.0081

[-2.78]***

[-1.43]

[0.20]

[-0.21]

D(Dependent Var. (-1))

0.4678

0.2801

1.0739

1.0850

[3.83]***

[1.70]

[9.44]***

[10.46]***

D(Dependent Var. (-2))

0.3732

0.2815

-0.6404

-0.641

[1.97]*

[1.28]

[-6.96]***

[-7.04]***

D(Dependent Var. (-3))

0.1612

0.0124

0.2004

0.2168

[0.88]

[0.04]

[4.85]***

[3.74]***

D(LrVaAv2(-1))

-1.1036

-0.7791

D(LrBALAv2(-1))

[-1.14]

[-0.78]

0.0272

0.0267

[1.59]

[1.39]

D(LrVaAv2(-2))

0.5667

0.6190

D(LrBALAv2(-2))

[0.53]

[0.53]

-0.0436

-0.0412

[-4.05]***

[-2.55]**

D(LrVaAv2(-3))

-0.5275

-0.2011

D(LrBALAv2(-3))

[-0.77]

[-0.24]

-0.0136

-0.0081

[-1.20]

[-0.59]

Constant

-0.2261

-0.2559

0.0081

0.0051

[-2.84]***

[-1.48]

[1.30]

[0.57]

R-squared

0.278

0.214

0.660

0.660

Adjusted R-squared

0.067

-0.016

0.573

0.573

Log likelihood

83.63

81.34

185.1

185.1

F-statistic

1.314

0.929

7.607

7.607

Akaike info criterion

-2.616

-2.531

-5.738

-5.738

Schwarz criterion

-2.137

-2.052

-5.285

-5.285

a) Pedroni Tests:

3.07E-03

2.96E-06

8.65E-07

1.62E-11

b) Wooldridge AR1:

0.0005

0.0030

0.0000

0.0000

c) Granger Prec. 4, (2) lags:

2.24E-11 (0.1432)

0.00E+00 (3.59E-13)

2.00E-04 (7.00E-07)

1.25E-05 (1.86E-05)

Notes: P-values in a)-c) are based on null hypotheses of: a) No panel cointegration, from estimates of Tables 6; b) No first-

order serial-correlation (Wooldridge AR test); and c) No Granger Precedence. For c), the first p-value tests if the

precedence.)

28

We now collect, in Table 11, evidence from these regressions on the dominance of Velocity and Balance

Elasticities for Registered and Non-Registered firms, respectively. From Lemma 3, Velocity – and not Balance

– elasticities dominate the countercyclical response of WIR Turnover for Registered firms: Since

and

, it follows that

. Thus our null hypothesis is

. Since the

alternative hypothesis is an inequality, it is evaluated by the one-tailed p-value. For these elasticities we can

take, say, the coefficients on Turnover, -0.525, in Table 7 (1b), and on Velocity, -1.871, in Table 8 (1b),

respectively. The Wald test on the null can now be rejected at p-values of 1.13E-04 and 6.31E-03, respectively.

For Non-Registered firms, by contrast, Balances should dominate the countercyclical elasticity of

Turnover:

with the null hypothesis that this difference should be zero. For these elasticities we

take the coefficients on first lagged term of Turnover, -0.519, from Table 9(1b), and that of Balances, -0.779,

from Table 10 (1b), respectively. The Wald test on this null can be rejected at p-values of 0.0399 or 0.3979.

This gives added support for our Lemma 3.

Table 11: P-Values: Wald Tests of Elasticity Dominance of Velocity and Balances over Turnover on

Value-Added Elasticities, Registered and Non-Registered WIR Clients

Registered

Null Hypothesis:

,

Alternative Hypothesis:

Non-Registered

Null Hypothesis:

Alternative Hypothesis:

FMOLS Estimates

DOLS Estimates

FMOLS Estimates

DOLS Estimates

7(1a):

-0.7861,

8(1a):

-0.8557.

7(1b):

-0.5249,

8(1b):

-1.8710.

9(1a):

-0.6702,

10(1a):

-1.1036.

9(1b):

-0.5192,

10(1b):

-0.7791.

p-values:

7(1a): 0.4366,

8(1a): 0.3509

p-values:

7(1b): 1.13E-04,

8(1b): 6.31E-03

p-values:

9(1a): 3.64E-03,

10(1a): 0.3279

p-values:

9(1b): 3.99E-02,

10(1b): 0.3979

By examining Table 11, we can confirm that the countercyclical dominance of R Velocity or NR

Balance predicted by Lemma 3 (here, the Alternative Hypothesis) does indeed hold in all cases. It is only

statistically significant, however, in 4 of the 8 Wald tests. Note also that significance was achieved 3 out of 4

times when the Wald test was calibrated on a Turnover regression, in Tables 7 and 9 – but only once when

based on a Velocity or Balance regression, in Tables 8 and 10.

29

VI. Estimates within Industrial Sectors

VI.1 Construction Sector

Table 12: Registered and Non-Registered Clients, CONSTRUCTION Sector: Log of Real WIR Turnover,

Velocity and Balances in CONSTRUCTION (LrTURN_Cons, LrVEL_Cons, LrBAL_Cons) Regressed on

Log of Real Value-Added in CONSTRUCTION, (LrVA_Cons)

t-statistics in ( ); ***: p-val < 0.01, ** : p-val < 0.05, *: p-val <0.10

Method: Vector Error Correction Model

Sample (adjusted): 1997 2008, Periods: 12

Sample (adjusted): 1997 2007, Periods: 11

(1a) Registered

(1b) Non-Registered

(2a) Registered

(2b) Non-Registered

COINTEGRATING

EQUATION

Depend. Variable:

Depend. Variable:

Depend. Variable:

Depend. Variable:

LrTURN_Cons(-1)

LrTURN_Cons(-1)

LrVEL_Cons(-1)

LrBAL_Cons(-1)

Variable

Coefficient

Coefficient

Coefficient

Coefficient

Constant

18.409

20.190

31.570

15.938

LrVA_Cons

-0.585

-0.771

-3.010

-0.425

[ -2.35]**

[ -4.88]***

[ -9.94]***

[ -2.59]***

VECTOR ERROR-

CORECTION EQUATION

Depend. Variable:

Depend. Variable:

Depend. Variable:

Depend. Variable:

D(LrTURN_Cons)

D(LrTURN_Cons)

D(LrVEL_Cons)

D(LrBAL_Cons)

Variable

Coefficient

Coefficient

Coefficient

Coefficient

Cointeg_Equa_RES(-1)

-0.543

-0.571

-0.120

-2.754

[-2.84]**

[-2.86]**

[-0.39]

[-1.67]

D(Dependent Var. (-1))

0.387

0.202

-0.052

1.180

[ 1.76]

[ 0.69]

[-0.09]

[ 0.79]

D(Dependent Var. (-2))

0.380

0.306

-0.037

-1.171

[ 1.45]

[ 1.04]

[-0.09]

[-0.58]

D(LrVA_Cons (-1))

-1.052

-0.760

-1.181

-2.981

[-3.14]***

[-1.85]

[-1.83]

[-1.55]

D(LrVA_Cons (-2))

0.538

0.206

0.310

0.393

[ 2.11]*

[ 0.62]

[ 0.31]

[ 0.24]

Constant

0.001

-0.014

-0.030

0.051

[ 0.10]

[-0.81]

[-0.91]

[ 0.42]

R-squared

0.777

0.718

0.562

0.826

Adj. R-squared

0.590

0.484

0.124

0.651

Sum sq. resids

0.005

0.007

0.021

0.113

S.E. equation

0.028

0.034

0.064

0.150

F-statistic

4.166

3.060

1.282

4.734

Log likelihood

29.867

27.757

18.923

9.592

Akaike AIC

-3.978

-3.626

-2.350

-0.653

Schwarz SC

-3.735

-3.384

-2.133

-0.436

a) Johansen-Fisher:

0.032

0.001

0.001

0.005

b) Lagrangian AR:

0.342

0.996

0.010

0.991

c) Granger Precedence:

0.027 (0.805)

0.002 (0.211)

0.109 (0.394)

0.407 (0.097)

serial-correlation with the lag-structure shown in this column; and c) No Granger Precedence. For c), the first p-value is on

the null that the independent variable does not Granger precede the dependent variable. (The p-value in parentheses is for

the reverse precedence.)

We will now explore the countercyclical patterns within industrial sectors. Comparing the elasticities of

Balances vs. Velocities between Registered and Non-Registered firms within a sector is of uncertain value here,

30

since there is no reason to believe that most WIR trade is intra-industry. A Registered manufacturing firm

seems at least as likely to trade with a Non-Registered services provider as it would with a Non-Registered

manufacturing firm, and so on. Nonetheless, substantial intra-industry trade may in fact exist, since we find a

similar pattern within most sectors.

Consider Construction, in Table 12 above. From Table 4, this is the sector with both the greatest

Turnover – almost a third of the Swiss total – and the most widespread acceptance: 37 percent of Swiss

construction firms accept WIR credits. It is possible that, due to the common practice of subcontracting, more

intra-industry trade takes place in Construction than in other sectors.

In Table 12 we are using annual Value-added, rather than its moving average as in Tables 3 and 7-10.

The significance of Table 12 suggests that Construction may be quicker to respond to cyclical trends than other

sectors. Most of the statistical tests here are similar to those earlier Tables, with the exception of that for

Autoregressive errors. Tables 7-10 are panel data, whereas Tables 3 and 12 are simple time series. Thus we can

use a Lagrange Multiplier test for serial-correlation. Here we cannot reject the null hypothesis of no serial-

correlation – giving us substantial confidence in the results.

Similarly to the panel results in Tables 7 through 10, we see in Table 12 that the first lagged Value-

Added term on Construction – in bold – has the countercyclical negative sign. Also similar to our panel results,

we see that for Registered firms, this Value-Added elasticity of Velocity in (2a) dominates the elasticity of

Turnover in (1a), in the sense of having greater absolute value. Balances (2b) similarly dominate Turnover (1b)

for Non-Registered firms. This Registered/Non-Registered pattern was shown in Table 11 to hold across all

our panel results. In what follows, we will refer to this as the standard countercyclical pattern.

In columns (1a) and (1b), note that Granger precedence flows from Construction Value-Added to WIR-

Turnover, but not in the opposite direction. In (2b), however, there is some evidence of precedence running in

the opposite direction, from WIR-Balances to Construction Value-added for Non-Registered firms. Again, this

may reflect the large size of some NR Construction firms (Winkler, 2010).

31

V1.2 Summary of Within-Sector Results: Contingency Tables

Sector-by-sector VEC regressions as shown above for Construction in Table 12 illustrate the ‘standard

countercyclical pattern.’ This pattern was not always replicated by each sectoral regression; it varies by

functional specification. Rather than choosing a ‘best’ specification for each sector, Table 13 summarizes

results for the following 4 regression forms:

i) One and two year lags of first-differenced variables in that sector;

ii) One and two year lags of first-differenced 2-year Moving Average of variables in that sector;

iii) As in a), but with a trend in the cointegrating equation; and

iv) As in b), but with a trend in the cointegrating equation.

Table 13 is a summary of contingency tables showing how often the following conditions were met: a)

the elasticity of WIR Turnover has the expected negative countercyclical sign and is significant, and b) either

the elasticity of Velocity dominates, or the elasticity of Balances dominates, as predicted by Lemma 3. We

count whenever these conditions are met for both Registered and Non-Registered firms. Note that there are 4

functional specifications for each relationship, and two lag structures for each independent variable. Thus there

could be as many as 8 instances for each of the 4 cells for the 6 sectors in Table 13, summarizing 8x4x6 = 192

regressions. But only the first or second lag is likely to be both of correct countercyclical sign and statistically

significant, so we should expect to see a maximum of 4 or 5 in most cells. This is so for all but the cell SUM,

which adds up all the other outcomes.

Note that we give the count for each of the 6 industrial sectors, plus the Unified merging all 6 sectors

into one, and finally the Sum counting up of All Effects separately. Our predicted ‘standard countercyclical

pattern’ is shown in a grey checker-board pattern of Velocity dominance for Registered and Balance dominance

for Non-Registered firms. This is so within the Construction, Services, Wholesale, and the Sum sections. Of the

sectors with enough observations to do a contingency test, only Manufacturing reverses this pattern of

countercyclical dominance.

Manufacturing shows countercyclical Velocity dominance among its Non-Registered, not its Registered

firms. We conjecture that the pattern of intra-sectoral purchases in this sector is likely to be the reverse of all

others. Rather than larger firms supplying output to – and accepting WIR currency from – the smaller ones, as

32

in most sectors, within Manufacturing it is common for smaller Registered firms to supply components for –

and accept WIR currency from – their larger Non-Registered counterparts. If this conjecture is correct, then this

reversal of our standard countercyclical pattern would be an ‘exception that proves the rule’ – because it

highlights the same structural logic.

Table 13: Countercyclical Dominance of WIR Balances (B) or Velocity (V),

Registered and Non-Registered Firms, Several Functional Specifications

Yates

Pearson

SUM,

All Effects

0.0856

0.0441

Reg

Non-Reg

Count_B=

6

11

Count_V=

15

7

CONST

0.6650

0.0833

RETAIL

NA

NA

Reg

Non-Reg

Reg

Non-Reg

Count_B=

0

1

Count_B=

0

0

Count_V=

2

0

Count_V=

2

1

HOSP

NA

NA

SERV

0.2357

0.0578

Reg

Non-Reg

Reg

Non-Reg

Count_B=

1

0

Count_B=

2

3

Count_V=

0

0

Count_V=

4

0

MANUF

0.2059

0.0350

WHOL

0.0528

0.0098

Reg

Non-Reg

Reg

Non-Reg

Count_B=

2

0

Count_B=

0

4

Count_V=

1

5

Count_V=

5

1

The p-values on the Pearson Chi-Squared two-tailed tests are given for each contingency table, along

with the Yates correction for continuity. Chi-Squares could not be calculated for the Retail or Hospitality

sector, since they had a zero row or column sum; these tables are greyed out. All of the Pearson tests are

significant at the 10% level. Given our small sample size, however Yates p-values may be more appropriate

here. These statistical results are therefore suggestive but not dispositive. Along with the regressions of Tables

33

7 through 12, however, they provide consistent evidence for the standard countercyclical pattern, as predicated

by Lemma 3.

VI. Conclusions

Previous work (Stodder, 2009) and updated regressions in Tables 2 and 3 show that WIR activity has

been highly countercyclical for 65 years in Switzerland. Tables 7-12 allow us to examine this countercyclical

trend over a shorter period, but within major industrial sectors. Regressions and Granger tests give evidence of

the ‘standard countercyclical pattern’ of WIR-exchange predicted by Lemma 3: Non-Registered firms supply

both product and credit to smaller Registered firms. Velocities of Registered firms, and Balances of the Non-

Registered firms play the leading countercyclical roles.

It is clear from Table 4 that WIR is a big part of the credit picture for SMEs, and also for some large

Non-Registered companies in Switzerland. But is the WIR-Bank an “only in Switzerland” case? Paradoxically,

the best evidence for the broader viability of WIR may be its very Swiss nature. WIR is spread across ethno-

linguistic regions, unlike other community-based cooperatives (Ostrom, 1990). German, French, and Italian-

speaking memberships are in rough proportion to their Swiss populations (WIR Rapport de Gestion, 2000-

2013). This suggests similar institutions could work elsewhere.

What about WIR’s inflationary potential? There is a broad literature (Mankiw, 1993; Mankiw and

Summers, 1986; Bernanke and Gertler, 1995; Gavin and Kydland, 1999) showing money supply is pro-cyclical.

Even less controversial is the finding that Velocity is pro-cyclical (Tobin, 1970; Goldberg and Thurston, 1977;

Leão 2005). By contrast, WIR Turnover is countercyclical, and our earlier VEC models (Stodder, 2009) show

that WIR Turnover and Swiss money supply are negatively correlated.

If WIR Turnover is countercyclical while that of national currency is pro-cyclical, then increases in the

former should be less inflationary than increases in the latter. Our estimates show that WIR is most used when

ordinary money is in shortest supply – during recessions, and where it is most needed – by SMEs. So WIR

Turnover is greatest when and where its inflationary potential is the least.

34

In addition to this non-inflationary bias, WIR activity may also ‘leverage’ far more economic activity

than suggested by its small size relative to Swiss money supply. Data for 2007, the most recent available to us,

show total WIR Balances of 612 million in SFr, only one-quarter of one percent of the basic Swiss money

supply, M1 (IMF, 2009). Consider, however:

The penetration of WIR into many sectors; e.g., 37 percent of all Swiss construction firms in Table 4.

Nearly twice as many Non-Registered as Registered firms in Table 4, including, as WIR statistician Winkler

(2010) notes, some that are quite well-known. The WIR activity of these large Non-Registered firms may

itself not be widely-known, however, due to its non-advertised nature.

Large Non-Registered firms show overall Turnover very close to that of smaller Registered firms in Table 4.

Lemma 3 and our regressions detail a structured reciprocity between large and small firms that is

fundamental to the countercyclical resiliency of WIR.

Large Non-Registered firms have smaller average WIR balances than Registered SMEs in Table 4. This

implies a higher ‘leveraging’ of Non-Registered SFr activity by WIR, as predicted in Lemma 2. Without

knowing the Swiss Franc expenditures of WIR clients, precluded by Swiss bank secrecy, we cannot estimate

the countercyclical ‘multiplier’ on WIR expenditures. The tiny ratio of WIR to M1, however, suggests that

it is likely be far higher than conventional multiplier estimates.

Lietaer et.al. (2009) argue that complementary currencies like WIR optimize a tradeoff between

efficiency and resiliency, mimicking a stable ecosystem. This contrasts with the impressive efficiency but

worrisome brittleness of our overall financial system. WIR resiliency is natural in the sense that the supply of

WIR-credits grows endogenously from trade, a trade that is itself countercyclical. As with a keystone species

within a stable ecosystem, the systemic resiliency gained by WIR’s leverage may be greater than is suggested

by its small “footprint.”

35

References

Arellano, M. 1987: Computing robust standard errors for within-groups estimators. Oxford Bulletin of

Economics and Statistics 49(4), 431-434.

Banerjee, A., and Carrion-i-Silvestre, J.L. 2006: Cointegration in panel data with breaks and cross-

section dependence. European Central Bank, Working Paper 591, February.

http://ssrn.com/abstract_id=885322, accessed May 1, 2016.

Berlin, M. 2003: Trade credit: why do production firms act as financial intermediaries? Federal Reserve

Bank of Philadelphia. Business Review 3rd Quarter, 21-28.

Bernanke, B., Gertler, M. 1995: Inside the black box: the credit channel of monetary policy

transmission. Journal of Economic Perspectives 9(4), 27-48.

Chauffour, J-P. and Malouche, M. 2011: Trade Finance during the Great Trade Collapse, World Bank

Publishing. http://elibrary.worldbank.org/doi/abs/10.1596/978-0-8213-8748-1, accessed May 1, 2016.

Clower, R. and Howitt, P. 1996: Taking markets seriously: groundwork for a post Walrasian

macroeconomics. In: Colander, (Ed.), Beyond micro-foundations: post Walrasian macroeconomics, Chapter 2.

Cambridge: Cambridge University Press.

Darity Jr., W., 1995: “Keynes’ political philosophy: the Gesell connection,” Eastern Economic Journal,

21(1), 27-41.

Defila, H., 1994. Sixty years of the WIR economic circle cooperative: origins and ideology of the

Wirtschaftsring. WIR Magazin. (Translated from the German by Thomas Geco.)

www.ex.ac.uk/~RDavies/arian/wir.html, accessed May 2, 2016.

Di Iorio, F., and Fachin, S., 2011: A panel cointegration study of the long-run relationship between

savings and Investments in the OECD economies, 1970-2007. Department of the Treasury of the Italian

Ministry of Economy & Finance, Working Paper No. 3, February.

http://www.dt.mef.gov.it/export/sites/sitodt/modules/documenti_it/analisi_progammazione/working_papers/DiI

orio-Fachin.pdf, accessed May 2, 2016.

Finnerty, J. 1980: Real money balances and the firm's production function: a note. Journal of Money,

Credit, &Banking 12(4), 666-671.

Fischer, S. 1974: Money and the production function. Economic Inquiry 12(4), 517-533.

Fischer, S. 1979: Capital accumulation on the transition path in a monetary optimizing model.

Econometrica 47(6), 433-1440.

Fisher, I. 1933: Stamp scrip. New York: Adelphi Co. Publishers. http://userpage.fu-

berlin.de/roehrigw/fisher/, accessed May 2, 2016.

Gavin, W.T. and Kydland, F.E. 1999: Endogenous money supply and the business cycle. Review of

Economic Dynamics 2, 347-369.

Goldberg, M. and Thurston, T.B. 1977: Monetarism, overshooting, and the pro-cyclical movement of

Velocity. Economic Inquiry 15(1), 26-32.

Gomez, G.M. 2008: Making markets: the institutional rise and decline of the Argentine red de trueque.

Hague: Institute of Social Studies.

Granger, C.W.J. 1988: Causality, cointegration, and control. Journal of Economic Dynamics and

Control. 12 (2–3), June–September, 551–55.

Greco, T.H., Jr. 2001: Money: understanding and creating alternatives to legal tender. White River

Junction, Vermont: Chelsea Green Publishing.

International Monetary Fund (IMF). 2009: International financial statistics.

http://www.imf.org/en/Data, accessed May 2, 2016.

Johnson, R.A. and Wichern, D.W. 1992: Applied Multivariate Statistical Analysis, 3rd Edition. New

York: Prentice Hall.

Keynes, J.M., 1964 [1936]: The general theory of employment, interest, and money. New York:

Harcourt Brace Jovanovich.

Klein, L. 1980: The Keynesian revolution, London: MacMillan.

36

Leão, P., 2005: Why does the Velocity of money move pro-cyclically? International Review of Applied

Economics, 19(1), 119-135.

Lietaer, B., Ulanowicz, R. E., Goerner, S. J., and Gomez, R. 2009: Quantifying sustainability: resilience,

efficiency and the return of information theory. Ecological Complexity 6(1), 27-36.

Linton, M. and Harris-Braun, E. 2007: LETSplay: Simulation of WIR-type system.

http://www.openmoney.org/letsplay/index.html, accessed May 2, 2016.

Love, I. 2011: Trade credit versus bank credit during financial crises, in Chauffour, J.P. and Malouche,

M. Trade finance during the great trade collapse, 25-40, World Bank.

http://elibrary.worldbank.org/doi/abs/10.1596/978-0-8213-8748-1, accessed May 1, 2016.

Lütkepohl, H. 1991: Introduction to multiple time series analysis, New York: Springer-Verlag.

Maddala, G.S., and Kim, I-M. 1998: Unit roots, cointegration, and structural change. Cambridge:

Cambridge University Press.

Maddala, G.S., and Wu, S. 1996: A comparative study of panel data unit root tests and a simplified test.

presented at Econometric Society Meetings, New Orleans, Jan. 1997.

Mankiw, N.G. (Ed.) 1993: Symposium on Keynesian economic theory today. Journal of Economic

Perspectives, 7, 3-4.

Mankiw, N.G. and Summers, L. 1986: Money demand and the effects of fiscal policies. Journal of

Money Credit and Banking 18(4), 415-429.

Nilsen, J. 2002: Trade credit and the bank lending channel. Journal of Money, Credit and Banking

34(1), 226-253.

Ostrom, E. 1990: Governing the commons: the evolution of institutions for collective action. New York:

Cambridge University Press.

Patinkin, D. 1956: Money, interest and prices: an integration of monetary and value theory. Evanston:

Row & Peterson.

Pedroni, P. 2001: Purchasing power parity tests in cointegrated panels. Review of Economics and

Statistics 83(4): 727–731.

Petersen, M. and Rajan, R. 1994: The benefits of lending relationships: evidence from small business

data. Journal of Finance, 49(1), 3-37.

______________________1997: Trade credit: theories and evidence. Review of Financial Studies 10(3),

661-691.

Rösl, G., 2006: Regional currencies in Germany: local competition for the euro? Economic Studies. 43.

http://www.bundesbank.de/Redaktion/EN/Downloads/Publications/Discussion_Paper_1/2006/2006_12_29_dkp

_43.pdf?__blob=publicationFile, accessed May 2, 2016.

Small Business Administration, 1998. Financing patterns of small firms: findings from the 1998 survey

of small business finance. Small Business Notes. http://www.smallbusinessnotes.com/small-business-

resources/financing-patterns-of-small-firms-findings-from-the-1998-survey-of-small-business-finance.html,

accessed May 2, 2016.

Stodder, J. 2009: Complementary credit networks and macro-economic stability: Switzerland’s

Wirtschaftsring. Journal of Economic Behavior and Organization 72(1), October, 79–95.

Studer, T. 1998: WIR in Unserer volkwirtschaft. Basel: WIR. Translation: Beard, P.H. 2006: WIR and

the Swiss National Economy. http://base.socioeco.org/docs/wir_and_the_swiss_national_economy.pdf, accessed

May 2, 2016.

Stutz, E. 1984: Le cercle économique-societé coopéreative WIR - une retrospective historique. Basle:

WIR. Terra, M.C., 2003: Credit constraints in Brazilian firms: evidence from panel data. Revista Brasilera de

Economia 57(2), 443-464.

Tobin, J., 1970: Money and income: post hoc ergo propter hoc? Quarterly Journal of Economics 84(2),

301-317.

Toda, H. Y and Yamamoto, T. 1995: Statistical inferences in vector autoregressions with possibly

integrated processes. Journal of Econometrics 66(1-2), 225-250.

37

Van Giersbergen, N.P.A, and Kiviet, J.F. 1994: How to implement bootstrap hypothesis testing in

regression models. Discussion paper TI 94-130, Tinbergen Institute, University of Amsterdam.

Wan, L., Riding, A., and Chamberlin, T. 2011: The global financial crisis: impacts on SMEs and public

policy responses. AGSE 8th International Entrepreneurship Research Exchange, Australian Graduate School of

Entrepreneurship, Swinburne University of Technology, Melbourne, Australia. February:

http://researchbank.swinburne.edu.au/vital/access/services/Download/swin:34256/SOURCE1, accessed May 2,

2016. White, Halbert. 1980: A heteroskedasticity-consistent covariance matrix and a direct test for

heteroskedasticity. Econometrica, 48(4), 817-838.

Williams, C.C. 1996: The new barter economy: an appraisal of local exchange and trading systems

(LETS). Journal of Public Policy, 16(1), 85-101.

Winkler, Stefan (WIR Statistician) 2010: Email correspondence, March 2010.

Winter-Ebmer, R., Zweimüller, J. 1999: Firm size wage differentials in Switzerland: evidence from job

changers. American Economic Review: Papers and Proceedings 89(2), 89-93.

WIR-Bank Panel Data, Balances and Turnover, 1994-2008: Courtesy of Mr. Stefan Winkler, Statistician

for the WIR-Bank, Basel, Switzerland.

WIR-plus magazine, various issues, 2000-13: Basel: WIR. Recent issues are online at

http://www.wir.ch/fr/medias/publications/#tab1, accessed May 2, 2016.

WIR-Bank, various years, 2000-13: Rapport de gestion, Basel: WIR. Recent years are online at

http://www.wir.ch/fr/medias/geschaeftsberichte/, accessed May 2, 2016.

Wooldridge, Jeffrey M. 2002: Econometric analysis of cross section and panel data, Cambridge, MA:

MIT Press.