Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed page of such transmission.

Mexico's Financial Sector Crisis: Propagative Linkages to Devaluation Author(s): Berry Wilson, Anthony Saunders, Gerard Caprio, Jr Source: The Economic Journal, Vol. 110, No. 460 (Jan., 2000), pp. 292-308 Published by: Blackwell Publishing for the Royal Economic Society Stable URL: http://www.jstor.org/stable/2565659 Accessed: 15/12/2008 12:03 Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the JSTOR archive only for your personal, non-commercial use. Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at http://www.jstor.org/action/showpublisher?publishercode=black. Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed page of such transmission. JSTOR is a not-for-profit organization founded in 1995 to build trusted digital archives for scholarship. We work with the scholarly community to preserve their work and the materials they rely upon, and to build a common research platform that promotes the discovery and use of these resources. For more information about JSTOR, please contact support@jstor.org. Royal Economic Society and Blackwell Publishing are collaborating with JSTOR to digitize, preserve and extend access to The Economic Journal. http://www.jstor.org

The Economic Journal, 110 (January), 292-308.? Royal Economic Society 2000. Published by Blackwell Publishers, 108 Cowley Road, Oxford OX4 1JF, UK and 350 Main Street, Malden, MA 02148, USA. MEXICO'S FINANCIAL SECTOR CRISIS: PROPAGATIVE LINKAGES TO DEVALUATION* Berry Wilson, Anthony Saunders and Gerard Caprio, Jr The sharp 1994 Mexican peso devaluation was followed by a financial-sector crisis, forcing the Mexican government to retake control of several banks and to grant substantial assistance to many others. This paper tests several hypotheses concerning the impact of devaluation. First, event-study methodology is used to test whether some sectors of Mexican economy were 'devaluation-gaining' while others were 'devaluation-losing'. Second, we test whether devaluation shocks were transmitted to the financial sector through the liability side versus the asset side of bank balance sheets. Our results indicate the importance of asset diversification. On December 20, 1994, Mexican authorities were forced to devalue the peso sharply, with consequences that included an economic downturn and a financial-sector crisis (Edwards, 1995). The devaluation occurred during a period of domestic economic growth, and just some two years after the privatisation of Mexico's financial sector in July 1992. The crisis forced the Mexican government to institute a rescue programme to retake control of some financial groups and to grant liberal assistance to many others. Several recent papers have investigated aspects of the Mexican peso and financial-sector crisis. Agenor and Masson (1999) analyse whether price movements in Mexico's government debt market anticipated the December 20, 1994 devaluation. Their study focused on the interest-rate spread between short-term peso-denominated and dollar-linked debt as a measure of devaluation risk. They find little evidence that Mexican debt markets anticipated devaluation prior to the December 20, 1994 event. Agenor and Masson (1999) argue that devaluation significantly weakened Mexico's financial sector by increasing the cost of servicing foreign-currency denominated debts. Similarly, Gruben and Welch (1996) pose a hypothesis that banks experienced a run of dollar-denominated bank deposits post devaluation, in anticipation that lender-of-last-resort guarantees would apply only to peso-denominated deposits. A scramble by depositors to move funds may have then generated a liquidity crisis in the wake of the peso devaluation. We term these bank-liability explanations of the devaluation-risk exposure faced by Mexican financial groups as the liability-exposure hypothesis.1 Alternatively, our study tests the hypothesis that the transmission of the peso-devaluation shocks occurred through the asset-side of bank balance sheets. Currency devaluations may often generate a differential impact across * The authors would like to thank the late Herb Baer for encouraging the research undertaken in the present paper, and Ross Levine and George von Furstenberg for helpful comments. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the World Bank, its Executive Directors, or the countries they represent. 1 Garber and Lall (1996) and Folkerts-Landau and Garber (1997) also discuss the role of derivative securities in exchange rate crises and macroeconomic stability. [ 292 1

[JANUARY 2000] MEXICO'S FINANCIAL-SECTOR CRISIS 293 sectors of an economy. Export-related sectors are often 'devaluation-gaining', while domestic/import sectors tend to be 'devaluation-losing'. If financial intermediaries have a significant net exposure, e.g., through bank loan portfolios, to devaluation-losing sectors over devaluation-gaining sectors, then devaluation shocks can become indirectly transmitted to the financial sector through the real sector. We term this asset-side transmission mechanism as the asset-exposure hypothesis. Our analysis is conducted in two phases. We first use event-study methodology to investigate the impact of the December 20, 1994 devaluation on the index returns (market-value changes) of six sectors of Mexico's real economy. In particular, we test whether some sectors were devaluation-gaining, while others were devaluation-losing. Like Agenor and Masson (1999), we also test whether the December 20, 1994 devaluation was anticipated by Mexican financial markets. However, our analysis uses equity-market data as opposed to their use of short-term Mexican debt-market data. We then test the two hypothesised transmission mechanisms using a multifactor asset-pricing approach with daily stock market data on Mexican financial-groups. Factors included in the analysis are five sectoral market indices, a broad Mexican market index dubbed the IPC (Index of Prices and Quotes), and relative changes in the peso exchange rate. Our analysis reveals that export sectors of Mexico's economy did indeed show strong economic gains post-devaluation, while construction and domestic sectors showed losses. Furthermore, the sectoral responses to devaluation appear to be important for explaining the financial-sector crisis that followed devaluation. Using an asset-pricing approach, we find that Mexican financial groups as a sector showed a strong sensitivity to devaluation-losing sectors, with virtually no sensitivity to devaluation-gaining sectors. Furthermore, little evidence was found of a direct link between exchange-rate volatility and financialsector performance, which would seem to contradict the liability-exposure hypothesis. One policy implication of the results is that greater diversification of asset risk across sectors of the economy would have significantly improved the post-devaluation performance of the financial-sector.2 Our results are consistent with those of other studies that link concentrated lending and insufficient diversification to financial-sector crises and bank failure. Regulation may impose limits on or disincentives to diversify (e.g., United States savings and loan regulations), 'connected' lending (Chile in the early 1980s), and the poor incentives for banks to plan and analyse their risk exposures with various banking systems. Although some developing countries might find it difficult to diversify due to their high output concentration in one or two sectors, Mexico would seem sufficiently large that this would not be the case.3 2 As discussed further in the text, the universal-banking powers of Mexican financial groups would have allowed them in principle to attain a significant degree of diversification through banking, brokerage, underwriting, insurance and leasing activities. Our analysis indicates that much of the devaluation risk was diversifiable. 3Other suggested contributing factors to Mexico's financial-sector crisis include undercapitalisation, poor supervision and inadequate banking expertise on the part of winning bidders in the privatisation.? Royal Economic Society 2000

294 THE ECONOMIC JOURNAL [JANUARY The paper is organised as follows. Section 1 provides background on the organisation of Mexico's financial sector, the post-devaluation financial-sector crisis and the government's rescue programme. Section 2 discusses data sources and Section 3 discusses the econometric approach and hypotheses tested. Section 4 then presents the results of the study's empirical analysis, followed by a discussion of our conclusions and policy implications in section 5. 1. Financial-sector Background 1.1. Mexico's Universal-bank Financial-sector Structure Privatisation of Mexico's banking sector was initiated under the Financial Groups Law, which became law onjuly 1, 1990. The law allowed private-sector majority ownership of Mexican banks. Individual or institutional ownership was limited to 5% equity or less (10% with government approval). Foreign investment was permitted up to a total of 30% of equity, but with a 5% cap on single individual or institutional ownership, and with industrial firms legally excluded from ownership.4 The 1990 Financial Groups Law adopted a universal-banking structure as the legal framework for organising Mexico's financial sector. Financial groups were headed by either a holding company, bank, or brokerage house, with most headed by brokerage houses. As well, each financial group was required to include at least 3 types of financial intermediaries, e.g., a bank, insurance company, leasing company, stock brokerage house, factoring or bonding concern, foreign exchange house, investment fund service, etc. (LatinFinance, 1/1/1993). Mexico's adopted universal-banking framework was considered one of the most advanced at the time (Martinez, 1993). Table 1 demonstrates the structure of a sample of Mexican financial groups, for whom data is available. Mexican financial groups began forming in 1991 to acquire privatised banks. Most of the 18 reprivatised banks were acquired by financial groups sponsored by brokerage houses, and all newly issued banking licenses were to newlyformed financial groups. In January 1994, the government divested its remaining minority stock holdings. This then completed the process of financialsector privatisation. 1.2. Banking-sector Signs of Weakness Signs of weakness in Mexican bank loan portfolios occurred as early as 1994. Post-privatisation, Mexican banks issued large numbers of credit cards, mortgages and car loans, partly in response to pentup consumer demand from previous periods of financial repression. In Mexico, past-due loans increased from 3.5% of total loans in 1991, to 8.5% in March, 1994, and then to 7.9% by 4 Privatisation was headed by the treasury ministry's (SHCP) bank privatisation committee.? Royal Economic Society 2000

2000] MEXICO'S FINANCIAL-SECTOR CRISIS 295 Table 1 Structure of Mexican Financial-Groups GFBANCOMER Subsidiaries Bancomer (commercial banking) Casa de Bolsa Bancomer (securities) Casa de Cambio Bancomer (exchange) Arrendadora Financiera Monterrey Almacenadora Bancomer Factoraje Bancomer Arrendadora Bancomer GFBANAMEX-ACCIVAL Subsidiaries Banco Nacional de Mexico Seguros Banamex Operadora de Sociedades de Inversion Banacci Acciones y Valores de Mexico Casa de Cambio Euromex Arrendadora Banamex Banamex Factoraje GF INVERLAT Subsidiaries Multibanco Comermex (commercial banking) Casa de Bolsa Inverlat (securities) Casa de Cambio Inverlat (exchange) Arrendadora Inverlat Factoring Inverlat Servicios Inverlat GFINVERMEXICO Subsidiaries Banco Mexicano InverMexico Casa de Bolsa Aseguradora InverMexico (investments) Arrendadora Financiera InverMexico (leasing) Factoring InverMexico Almacenadora InverMexico-USCO (51%) (warehousing) Afianzadora InverMexico (bonding) InverMexico Household Tareta de Credito, Sociedad Financiero de Objeto Limitado (credit card) LatInvest Securities, LTD (United Kingdom) LatInvest Securities, LTD (United States) GFPROBURSA Subsidiaries Multibanco Mercantil Probursa Seguros Probursa Probursa Casa de Cambio Probursa Fianzas Probursa Factoraje Probursa Almacenadora Probursa Arrendadora Probursa Promotora Probursa GF SERFIN Subsidiaries Banca Serfin (commercial banking) Operadora de Bolsa Serfin (brokerage) Factoraje Serfin (factoring) Arrendadora Serfin (leasing) Almacenadora Serfin (warehousing) Afinazadora Insurgentes (bond operations) Seguros Serfin (insurance) Servicios Corporativos Serfin? Royal Economic Society 2000

296 THE ECONOMIC JOURNAL [JANUARY MULTI-VA GFsubsidiar-ies Banco del Centro, S.A. MultiValores Casa de Bolsa MultiVa Factoring MultiVa Arrendadora Casa de Cambio Amercam Source: Moody's International Manual. year-end 1994, representing an increase from 35.1% to 97.5% of total bankingsector book capital (IMF, 1995, p. 125).5 Overdue loans became a national epidemic after the sharp currency devaluation on December 20, 1994. Borrowers defaulted on loans because of high interest rates and the post-devaluation economic decline. Interest rates on consumer debt peaked as high as 120%, and then subsided to about 78% in 1995. At Bancomer, one-third of all customers had to reschedule loan payments, and a smaller percentage defaulted. Many Mexican banks also had trouble meeting capital requirements in 1992. Nine banks failed to meet the 8% (BIS) capital requirement by the December 1992 deadline. This included four of Mexico's six largest banks (Serfin, Multibanco, Comermex and Banco Internacional) with capital levels below 6%, and smaller banks including Banco Mexicano, Banco BCH, and Banpais with capital ratios at 6% to 7%, and finally Banco de Oriente, Banca Cremi, and Banco del Atlantico with capital ratios above 7%. The Finance Ministry was forced to grant extensions until October 1993 (Institutional Investor, 1993). Many financial groups were also reported to have suffered large trading losses in 1993 on ajustabonos (three- and five-year inflation-adjusted government bonds), with Probursa losing approximately one-third of the amount paid by its owners for its bank, Multibanco Mercantil, in the privatisation. Total financial-sector losses from ajustabonos were reported at $1 billion. The losses were said to reflect the embryonic state of asset-liability management of Mexican banks (Institutional Investor, 1993). Mexican reporting standards also may have contributed to an under-reporting of the severity of loan default rates, by requiring only past-due interest but not past-due principal amounts to be reported. Bad loan estimates in March, 1996 ranged from 15% to 40% of the banking sector portfolio, with a bailout price tag ranging from $10 billion to $30 billion, or from 5% to 12% of GDP (Latinfin, 3/1/96).6 New accounting standards for banks are currently being implemented (Garcia-Cantera and Burbridge, 1997). 1.3. The Post-Devaluation Bank Rescue Plan In response to mounting bank loan losses, the state-owned deposit insurance agency, FOBAPROA (Fondo Nacional de Proteccion al Ahorro), developed a 5 De la Cuadra and Valdes-Prieto (1992) describe a similar history of events from financial repression to rapid loan expansion in Chile. 6 In comparison, a United States bank is considered unhealthy if past due loans are more than 3% of total loans.? Royal Economic Society 2000

2000] MEXICO'S FINANCIAL-SECTOR CRISIS 297 plan to recapitalise banks by purchasing non-performing loans, according to a '2 for 1 plan'. FOBAPROA proposed to buy $2 worth of non-performing loans for every $1 of new capital injected. The government's loan purchase programme has included the following transactions: In September 1995, GF GBM-Atlantico and GF Promex-Finamex sold more than US$1.06 billion in overdue loans to FOBAPROA (Sourcemex, 9/13/1995). In June 1995, the Mexican government agreed to buy $783 million of debts from Probursa and Serfin (The Banker, 7/5/1995). The Probursa sale facilitated a majority share sale of Probursa stock to Banco Bilbao Vizcaya. GF Serfin SA sold more $700 MM in loans to the government in June 1995 and planned to sell $1 billion more (Wall StreetJournal, 1/25/96). The Mexican government issued 10-year bonds to Banamex in exchange for US$2 billion in Banamex loans, acquired by FOBAPROA. 43% of the loans were already past due (The Banker, 2/1/1996). Banco Mexicano sold US$900 million in loans to Fobaproa in late 1995 (The Banker, 2/1/1996). Mexican banking authorities also seized control of several banks and credit unions, beginning with the seizure of Grupo Financiero (GF) Cremi-Union SA. Then, in March 1995 authorities seized control of GF Asemex-Banpais and its banking arm, Banpais, the country's ninth largest bank. The government effectively took over GF Inverlat when it was unable to roll over short-term dollar debt in August, 1995. The Bank of Nova Scotia acquired 8% of the financial group (Wall StreetJournal, 12/1/1995). Fraud also played some role in the banking crisis. In particular, In May 1994, the SHCP took control of Grupo Havre, under accusations of fraudulent insider loans linked to the government's small business lender, Nacional Financiera (NAFINSA), and the Foreign Trade Bank (BANCOMEXT) (SourceMex, 9/14/1994). Investigators uncovered $700 million in insider loans with GF Cremi- Union. The bank's chairman was sought as a fugitive, and Mexican regulators took control on September 6, 1994. (Houston Chronicle, 9/7/1994). Regulators found that 60% of Banco Interestatal's loans were internal loans to the bank's controlling group. (Emerging Markets Report, 9/21/95). Finally, the banking crisis also attracted foreign-bank entry. In particular, In May 1995, Banco Bilbao Vizcaya increased its stake in GF Probursa from 22% to 70%, paying $390 million on condition that Fobraproa take on $780 million of loans (LatinFinance, 3/1/1996). In August 1995, Banco Central Hispanoamericano and Banco Comercial Portugues made similar deals with Banco Internacional (LatinFinance, 3/1/1996). Bank of Montreal agreed to pay between $426 million and $480 million? Royal Economic Society 2000

298 THE ECONOMIC JOURNAL [JANUARY for a 16% stake in Bancomer (LatinFinance, 3/1/1996). Scotiabank increased its stake in GF Inverlat to 55%, costing $225 million (LatinFinance, 3/1/1996). Total cost of the bank rescue efforts have been put at around $65 billion (Euromoney, 9/10/98). However, the Mexican Congress has recently passed legislation which may reduce these costs. The new law replaces Fobaproa with a new institution, called the Institute for the Protection of Bank Savings. The new body may increase risk-sharing of Fobaproa losses with banks to reduce the cost of the bank bailout. The new legislation will also allow majority ownership of banks by foreign investors for the first time (U.S. Newswire, 12/24/98) and will cap deposit insurance guarantees (initially set at $2 million per depositor) with the objective of eventually reducing the coverage level to same as that in the U.S. (i.e., 100,000). 2. Data Daily data on (i) the peso/dollar exchange rate, (ii) a broad Mexican-market index (the IPC) and six Mexican sectoral market indices and (iii) stock price data on Mexican financial groups were drawn from El Financiero, a leading Mexican financial newspaper. The market index and the six sectoral indices: the industrial/extractive, communication/transportation, service, commercial, construction and transformation, are constructed by the Mexican stock exchange. The present analysis focuses on ten Mexican financial groups with sufficient trading activity over the study period to include in the present analysis. The study focuses on financial groups as a whole, rather than on their banking, brokerage or insurance subsidiaries, since the most active trading is in the parent stock. In cases where there are multiple classes of parent stock that trade, we chose that class which appeared to be the most active. Summary statistics for the study variables are presented in Table 2, both for the entire period 1993-6 and by individual year. Also, Fig. 1 plots the course of Table 2 Daily Rate of Return Statistics - % Variable 1993 1994 1995 1996 1993-6 Peso Return -0.01 0.40 0.25 0.00 0.14 Broad-Market Returns 0.21 0.03 0.34 0.18 0.21 Sample GF Returns 0.40 0.30-0.52 0.01 0.07 Sectoral-Index Returns Commercial 0.21-0.10 0.02 0.14 0.07 Communication/Transportation 0.23 0.00 0.13 0.02 0.09 Construction 0.35-0.06 0.13 0.09 0.12 Industrial/Extractive 0.24 0.48 0.54-0.06 0.62 Service 0.12-0.32-0.08 0.08-0.04 Transformation 0.31-0.11 0.22 0.12 0.14?) Royal Economic Society 2000

2000] MEXICO'S FINANCIAL-SECTOR CRISIS 299 Peso[U.S.$ 7.50 7.00 6.50 6.00 5.50 5.00 Peso exchange rate 4.50 4.00 3.50 ON ON ON ON ON ON ON ON ON ONl ON, ON, ON, ON, ON, ON ON ON ON ON ON ON ON ON ON ON ON ON ON ON, ON, - - - - - -~ -~- - C, a, a-, - -?ClC) Cl Cl l -, --, r- Cl0 Cl Fig. 1. Peso/dollar Exchange Rates over the Event Period the peso over this period. The sharp increase in the peso/dollar exchange rate on December 20, 1994 is the 'devaluation event' of interest to this study. The event represents a regime shift from a pegged to a floating exchange rate, in part a consequence of the near-depletion of Mexico's foreign currency reserves at the time. Finally, note that the peso exchange rate continued to devalue post-event date, and became more volatile. Figs 2-4 plot the six Mexicanmarket sectoral indices of interest. 140 1 30 12oa0 cooi ocey20 Fig. ~~~Mining 2. adcmuiaintasotto nie

300 THE ECONOMIC JOURNAL [JANUARY 110-100- _ 90-80- 70-60 t Commercial 50 I i i l i i i l 10/29/94 11/18/94 12/8/94 12/28/94 1/17/95 2/6/95 2/26/95 3/18/95 4/7/95 4/7/95 5/17/95 Fig. 3. Manufacturing and Commercial Sector Indices 105-K 95 - Construction 85 75 65 55 Service sector 45 35 1 1 1 1 I- l I 10/29/94 11/18/94 12/8/94 12/28/94 1/17/95 2/6/95 2/26/95 3/18/95 4/7/95 4/27/95 5/17/95 Fig. 4. Construction and Service-sector Indices 3. Econometric Tests We divide our analysis into two parts. First, we use event-study methodology to test whether sectors of the Mexican economy showed a differential response to the December 20, 1994 devaluation event. Second, the analysis uses an assetpricing approach to test the linkage mechanism between the devaluation event and the financial-sector crisis that followed.?) Royal Economic Society 2000

2000] MEXICO'S FINANCIAL-SECTOR CRISIS 301 3.1. Event-study Tests One differential response to devaluation would be that export-sectors of the economy show gains post devaluation, while domestic/import sectors show losses. Changes in stock-market valuations will reflect investors' expectations of the economic impact of devaluation on sectors of Mexico's traded economy. Accordingly, we use a three-day event window (December 20-22, 1994) to test for devaluation effects. A significant valuation response to devaluation would signify that investors did not anticipate the devaluation event. The analysis also incorporates an event window covering the thirty days prior to devaluation to capture potential leakage of information pre-devaluation. In this case, insiders may have had access to valuable devaluation-related information prior to the 12/24/94 event. Therefore, this window tests for event leakage. The analysis also includes the rate of change in the peso exchange rate itself as a variable, to control for the impact of peso volatility. 3.2. Asset-pricing Analysis of the Financial-sector Crisis Next, the study uses an asset-pricing approach to analyse the potential influence of peso volatility (devaluation) and other factors on the stock-market returns of the Mexican financial groups. To test the liability-exposure hypothesis, we include the peso exchange rate itself as one factor in the analysis. If financial groups had a direct exposure to peso volatility through potential valuation effects on the liability side of their balance sheets, then the peso exchange rate factor should be a significant explanatory factor of financialgroup returns. In contrast, under the asset-exposure hypothesis, financial groups were affected by devaluation through the asset side of their balance sheets, due to equity or debt-like (e.g., lending) exposures to sectors of the Mexican economy. In this case, including the sectoral indices as factors should reveal which sectors the financial groups were sensitive to, which would then determine their response to devaluation. We use the 1993-5 period as the study interval for the asset-pricing regressions, and then include the year 1996 as a robustness test of the results. 4. Results 4.1. Impact of Devaluation on Mexico's Real Sectors Table 3 presents the event-study results. The event-window dummy variables are the three-day event window dummy (DEVAL) covering December 20-22, 1994 and the thirty-day pre-devaluation event-window dummy (PRJOR&30). The variable, PESO, is the rate of change of the peso exchange rate. Estimation covers the period: 11/20/94 to 1/4/95. Finally, while the service-sector index includes both real and financial services (including financial-group performance), we nonetheless include it here for completeness. First, the 3-day event dummy (DEVAL) tests for direct devaluation effects. (? Royal Economic Society 2000

302 THE ECONOMIC JOURNAL [JANUARY Table 3 Impact of Devaluation on Mexico's Real Sector Sector index R N DEVAL PRIOR30 PESO Commercial 1.2 712 1.36-0.468-0.0465** Communication/Transportation 3.7 712 5.20 * -0.212-0.0680*** Construction 0.7 712-0.623-0.541-0.0531** Industrial/Extractive 7.8 711 9.16*** -1.09** -0.0373 Service 2.9 712-3.90*** -0.305-0.0264 Transformation 0.8 712 0.821-0.488-0.0303* **, *** significant at 10%, 5% and 1 % levels respectively. The results indicate that two sectors, namely the industrial/extractive and communication/transportation sectors, showed significant positive devaluation 'surprises' of 9.16% and 5.20%, respectively. In contrast, the service-sector index showed a significant negative response (declining -3.90%) to devaluation. Finally, the commercial and transformation sectors showed positive responses, and the construction sector showed a negative response, although these were not significant. Next, the prior thirty-day window dummy, PRJOR&30, estimates proved not to be significant (except with the industrial/extractive sector) and always showed a negative sign, in contrast to the differential response observed for the devaluation-event window. In other words, the coefficient estimates do not display the export/gain and import/lose pattern as with the 3-day window. Therefore, the results appear to confirm those of Agenor and Masson (1999), that the financial markets did not anticipate the devaluation event, prior to the December 20, 1994 date. Finally, the PESO variable coefficients all display negative signs, and are significant in five of seven cases. These financial-market results indicate that investors reacted to devaluation with the anticipation that export sectors would show gains post-devaluation, while domestic consumer sector would show losses. The results also indicate that investors did not anticipate the devaluation event. 4.2. Asset-Pricing Analysis of Mexico's Financial Groups We next test the two hypothesised linkages between the devaluation crisis and Mexico's financial-sector, namely the liability-exposure hypothesis and the asset-exposure hypothesis. The analysis uses a fixed-effects approach to simultaneously estimate separate financial-group sensitivities (Betas) with each particular factor. For example, a factor model with the five sectoral indices and separate Beta estimates for each of the ten financial groups would include 50 factors in a full model representation. This framework is then used to test the significance of any particular factor. First, partial F-statistics are calculated from a full model that includes the ten fixed-effects terms for that particular factor, versus a reduced model regression that excludes the ten terms. The partial F-statistics are then compared with (? Royal Economic Society 2000

2000] MEXICO'S FINANCIAL-SECTOR CRISIS 303 critical values to determine significance of the factor's impact on the financialgroup sector as a whole. Insignificant factors can then be eliminated from further consideration. This approach serves to minimise the possibility of obtaining spuriously-significant results for some subset of the financial groups. Tables 4 and 5 summarise the results. Panel A of Table 4 reports the first set of elimination statistics, where the initial fiill model includes the five sectoral indices.7 Since at this stage our interest is to isolate those economic sectors that appear to have significantly influenced financial-group returns, panel A presentsjust the partial F-statistics. The columns labelled 1-4 gives the partial F-statistics from the sequential elimination tests of the sectoral indices in turn. The '*' indicates cases where the resulting F-statistic exceeds its critical value. On the first round, the A. Sectoral-index Elimination Rounds Table 4 Portfolio Decomposition Results Round - elimination F-statistics Sectoral Index 1 2 3 4 (1) Commercial 1.31 - - - (2) Communication/ 5.87* 5.51* 5.36* 4.33 Transportation (3) Construction 8.33* 10.14* 10.32* 44.01* (4) Industrial/Extractive 2.48 2.49 - - (5) Transformation 3.31 4.36 4.24* - B. Mexican market-index and peso return Elimination Rounds Round - Elimination F-Statistics Index 1 2 3 4 (1) Mex. Index: IPC 0.58 - - - (2) Peso Return 2.47 2.51 2.47 (3) Fixed-Effect Intercepts 1.96 1.94 - C. Consumer-index & Sears-index Elimination Rounds Round - Elimination F-Statistics Index 1 2 (1) Consumer Index 3.89 - (2) Sears Index 9.40* D. Event-window Elimination Rounds Round - Elimination F-Statistics Index 1 2 (1) Pre-Deval Fixed Effects 0.68 - (2) Deval Window 1.27 (3) Post-Deval Fixed Effects 2.44 7 This decomposition can be thought of as deriving a set of 'weights' that represents the portfolio allocation of financial-group assets to each sector of the economy. Within the universal-banking framework, the financial groups would have in theory the ability to take equity-like and debt-like positions in different sectors. (D Royal Economic Society 2000

304 THE ECONOMIC JOURNAL [JANUARY commercial-sector index produces the smallest (non-significant) F-statistic, and it is therefore eliminated from future rounds. In turn, three additional sectoral indices are eliminated in this fashion, namely the industrial/extractive (round 2), the transformation (round 3) and the communication/transportation (round 4) sectors. Therefore, of the five sectoral indices, the financial groups appear to show sensitivity mostly to the construction sector. In Panel B of Table 4, an alternative factor structure is investigated, which includes the IPC (the broad Mexican-market index), the peso factor and a set of fixed-effect intercepts which allow separate 'alphas' to be estimated for each financial group. The results reported in panel B indicate that none of the factors in this alternative factor structure proved significant. In panel C of Table 4, two additional 'sectoral' indices are introduced into the analysis. As noted above, the service-sector index is comprised of both real services and financial services, and therefore already includes financial-group performance. To construct a purely real-sector index that might reflect the impact of devaluation on the consumer sector, we attempt two approaches. First, we construct a sectoral index from daily stock-market data on Mexican retail companies, which is labelled 'Retail-Index' in Table 4. However, many of these tend to be high-end retail outlets that may not well represent the impact of devaluation on middle-class Mexican consumers. Therefore, we construct a separate index of daily stock returns from just the Sears' Mexican subsidiary, labelled 'Sears Index' in Table 4. As in the United States, this retail company appears to better reflect Mexico's middle-class than the other retail outlets. The results listed in panel C demonstrate that both the consumer-sector index and the Sears index show large positive relationships to financial-group performance. However, only the Sears index partial F-statistic exceeds the critical value. Finally, panel D of Table 4 investigates the event impact of devaluation within the context of the present analysis, i.e., here covering the entire period 1993-5. Three windows are constructed. First, the 'Deval Window' is as defined before. Second, the variable, 'Pre-Deval Fixed Effects' is a thirty-day pre-devaluation window, with separate effects for each financial group. Finally, the variable, 'Post-Deval Fixed Effects', is a thirty-day window following the three-day devaluation window, again with separate effects for each financial group. Note that none of these factors proved significant. The results presented in Table 4 indicate that the construction-sector index and the specially-constructed 'Sears' index are the most promising factors. Therefore, Table 5 pursues a two-factor model that includes these two factors. Panel A of Table 5 lists the results of the analysis. The overall R2 is 14.3%. The column labelled 'a %' gives the estimated alphas for each financial group (estimated as fixed-effect intercepts). The alpha estimates are generally negative but not statistically significant. The columns labelled 'Construction Index' and 'Sears Index' give the factor-sensitivity estimates of the two factors, respectively. The factor-sensitivity estimates are always positive when significant. Panel B of Table 5 investigates a further issue with the estimated factor (? Royal Economic Society 2000

2000] MEXICO'S FINANCIAL-SECTOR CRISIS 305 Table 5 Multi-index Regressions - 1993-95: Mexican Financial Groups A. Two-Index Fixed-Effects Model R2 = 14.3% Financial Group a% Construction index Sears index BANACCIA -0.029 0.408*** 0.263*** GBMATLA A -0.100 0.051 0.022 GFBA -0.185 0.671*** 0.194*** GFINBURA 0.252 0.191*** 0.082* GFINLATA -0.360** -0.010-0.032 GFINVER A -0.189 0.024-0.037 GFNORTEB -0.254 0.263*** 0.153*** GFPROBUB -0.294** 0.094** 0.375*** GPROHNB -0.059-0.046 0.051 GSERFINA -0.201-0.007 0.103** B. Up/down Construction Index Betas R2 = 14.8% Construction index Financial group a% Down beta Up beta Sears index BANACCIA -0.112 0.40*** 0.40*** 0.271*** GBMATLA A -0.234 0.14* -0.02 0.038 GFB A 0.104 0.68*** 0.66*** 0.198*** GFINBURA 0.069 0.16* 0.21*** 0.080* GFINLATA 0.182 0.33*** -0.24*** -0.026 GFINVER A -0.013 0.15* -0.07-0.026 GFNORTEB -0.696** 0.11 0.35*** 0.157*** GFPROBUB -0.058 0.31*** -0.04 0.367*** GPROFINB -0.075 0.02-0.09 0.051 GSERFINA -0.022 0.12-0.10 0.115** * significant at the 10% level. ** significant at the 5% level. * significant at the 1% level. sensitivities. Investors may have anticipated that the Mexican government would give financial ('bail out') support to financial groups in financial distress. In particular, this support would have tended to diminish the sensitivity of returns in down markets. Accordingly, Panel B of Table 5 reports separate 'up beta' and 'down beta' construction-sector sensitivity estimates for each financial group, while continuing to include the Sears index. The resulting estimates suggest that the financial group returns were generally equally sensitive to both up and down movements in the construction-sector index. Finally, as a sensitivity test for the sample period used in the preceding analysis, results are presented in Table 6 which updates the regression analysis to include the year 1996. In general, only small changes are observed to occur in the parameter estimates, indicating that the previous results are robust and that a similar return generating process continues to hold when the 1996 data are included. In summary, the asset-pricing regressions give reasonably strong evidence in favour of the asset-exposure hypothesis. Little evidence was found of a direct link between devaluation and financial-sector performance. Inclusion of fac- ( Royal Economic Society 2000

306 THE ECONOMIC JOURNAL [JANUARY Mexican Financial Groups Table 6 Multi-index Regressions - 1993-96: R2 = 12.1% Financial group a% Construction index Sears Index BANACCIA -0.006 0.403 0.233*** GBMATLA A -0.057 0.047-0.003 GFB A -0.101 0.701*** 0.167*** GFINBUR A 0.204* 0.191*** 0.074** GFINLATA -0.306** -0.010-0.030 GFINVER A -0.244** 0.029-0.063* GFNORTEB -0.254 0.263*** 0.153*** GFPROBUB -0.116 0.131*** 0.352*** GPROHNB 0.018-0.063 0.061 * GSERFINA -0.204* 0.018 0.081** * significant at the 10% level. ** significant at the 5% level. * significant at the 1% level. tors including the peso exchange rate and the devaluation event window proved insignificant. In contrast, the financial groups did show significant factor sensitivities to the construction sector index and the Sears index, but no significant factor sensitivities to export-related sectors. These results are further corroborated by the aggregate sectoral loan statistics presented in Table 7, where loan percentages are high for domestic sectors such as commercial and construction, while low for export sectors such as mining. The implication is that the financial-sector problems in Mexico have dragged on so Table 7 Mexican Bank's Aggregate Sectoral-Loan Statistics Million pesos Percentages Sector 1994 1995 1996 1994 1995 1996 (1) Agriculture 31,713.7 29,963.6 35,980.5 6.21 4.65 5.08 (2) Mining 3,041.5 2,347 2,395 0.60 0.36 0.34 (3) Manufacturing 90,608 108,848.3 196,583.5 17.74 16.89 27.73 (4) Construction 77,450.8 93,632.6 107,228.9 15.17 14.53 15.12 (5) Commercial, 192,685.4 231,196.1 213,240.3 37.73 35.88 30.08 restaurants and hotels (6) Real services, 58,851.3 68,555 74,277.9 11.52 10.64 10.48 communications, transportation (7) Financial Service 19,985.6 20,694.3 13,640.2 3.91 3.21 1.92 (8) Government 17,375.4 32,076.6 35,874.3 3.40 4.98 5.06 (9) External 8,273.2 11,033 12,818.6 1.62 1.71 1.81 (10) Other 0 35,620.5 0 0.00 5.53 0.00 (11) Interbank Credit 10,726.9 20,475.6 16,935.9 2.10 3.18 2.39 Source: Central Bank of Mexico. (? Royal Economic Society 2000

2000] MEXICO'S FINANCIAL-SECTOR CRISIS 307 long, and have been so difficult to reverse, because the sectors of the economy they were most involved with have shown the slowest recoveries. 5. Conclusions Our analysis yields the following conclusions. First, sectors of the Mexican economy showed significant differential responses to devaluation. Export sectors tended to gain, while domestic consumer-related sectors tended to decline. The significant surprise effects at devaluation indicate that the possibility of a peso devaluation was not anticipated by investors prior to the devaluation event. Second, our analysis tests whether Mexico's universally-organised financial groups were able to diversify across sectors that showed differential valuation effects post-devaluation. Our results demonstrate that their performance was strongly tied to the domestic consumer-sector, which showed the largest losses and slowest recoveries post-devaluation. This conclusion appears to be consistent with both the depth and longevity of the financial-sector crisis. Our results indicate that a lack of diversification was a contributing factor to the banking crisis. Third, one implication of our results for Mexico is that the banking sector will not recover from the crisis until, among other things, the consumer-loan exposures of financial groups are reduced, or the consumer sector shows a strong recovery. Indeed, the approach taken by the Mexican government to recapitalising the financial sector underscores our conclusion of asset-side problems. As discussed above, Fobaproa's approach had been to reduce loan exposures by purchasing non-performing loans from financial groups. In contrast, liability-side exposures would be more easily reduced through a reorganisation of liability and equity claims, much as with a Chapter 11 reorganisation of a U.S. firm. Finally, it is interesting that following privatisation, financial groups expanded their loan portfolios mostly along the dimension of the consumer sector. However, this may have been a consequence of several factors. First, export-sector (and perhaps many other established public) companies may have traditionally relied on domestic and foreign stock and bond markets for raising capital, particularly during the period when the banking sector was nationalised. Second, consumers may have responded to the period's liberalisation with a great deal of pent-up demand for consumer credit, creating growth opportunities for banks. Relatedly, perceived profit margins from the consumer sector may have been large relative to other sectors prior to devaluation. Federal Communications Commission New York University Thae World Bank Date of receipt offirst typescript: November 1997 Date of submission offinal typescript:july 1999 (? Royal Economic Society 2000

308 THE ECONOMIC JOURNAL [JANUARY 2000] References Agenor, P. R. and Masson, P. R. (1999). 'Credibility, reputation, and the Mexican peso crisis,' Journal of Money, Credit and Banking, vol. 31 (1), pp. 70-84. De la Cuadra, S. and Valdes-Prieto, S. (1992). 'Bank structure in Chile,' in (by G. G. Kaufman ed.) Banking Structures in Major Countries, Amsterdam: Kluwer. Edwards, S. (1995). Crisis and Reform in Latin America: From Despair to Hope, Oxford: Oxford University Press. Folkerts-Landau, D. and Garber, P. M. (1997). 'Derivative markets and financial system soundness,' Mimeo. Garber, P. M. and Lall, S. (1996). 'Derivative products in exchange rate crises,' Mimeo. Garcia-Cantera, J. A. and Burbridge, P. (1997). 'Mexican banking system: final definition of new accounting rules,' Salomon Brothers. Gruben, W. C. and Welch, J. H. (1996). 'Default risk and dollarization in Mexico,' Journal of Money, Credit and Banking, vol. 28(3), pp. 393-401. IMF (1995). International Capital Markets: Developments, Prospects, and Policy Issues, Washington, D.C. Institutional Investor, (1993). 'Land of the giants', vol. 27, no. 2. Martinez, G. 0. (1993). 'The Modernization of the Mexican financial system,' in (Shakil Faruqi ed.) Financial Sector Reforms in Asian and Latin American Countries: Lessons of Comparative Experience, Economic Development Institute of The World Bank. (D Royal Economic Society 2000