Terrorizing Consumers and Investors Samer Haj-Yehia* Analysis Group, 111 Huntington Ave., Boston, MA 02199, USA January 2006 Abstract This paper provides evidence on how consumers and investors react to terror attacks based on a new database from the Israeli-Palestinian conflict. An increase in terror casualties triggers households to alter their perceived personal security and expected future income. Only ex-post do households distinguish a temporary from a permanent increase in terror casualties. A temporary increase in the number of terror casualties causes a bust-boom cycle of durables consumption and irreversible investment; nondurables are affected less. A permanent increase in the number of terror casualties causes a one-time drop in consumption. This is in line with the theory on irreversible investment and durables consumption: terror generates temporary uncertainty about personal security and future income, which in turn causes a bust-boom cycle of durables due to bunching of purchases in later periods. A permanent increase in terror causes neither bunching nor boom. Similar results are obtained for the effect of terror casualties on fixed capital. Keywords: Consumption; Investment; Uncertainty; Terror; Israel; Durables; Nondurables. JEL classification: E21; E27; G31; Z00 * Email address: samer_hajyehia@ 1
,* 11. Introduction While terror is not new, the effect of terror on consumption and investment has only recently been empirically examined in economic studies. A major effect of terror is the creation of 2uncertainty vis-à-vis personal security as well as future income. Against this background, this paper addresses the following questions: (1) What are the impulse response functions of consumption and investment? (2) Are the impact response functions for durables, nondurables, and services different? (3) Is there a long-term effect of terror on consumption and investment? (4) Do consumers develop a hedonic adaptation to terror? Answers will assist policymakers to understand how terror affects consumption and investment, which together constitute significant portions of GDP, and perhaps to neutralize the effects of terror in a timely fashion to prevent destabilization of the economy. The above questions are addressed using a new database on Israeli consumption and terror casualties from the Israeli-Palestinian conflict for the period 1980-2002. As Figure 1 shows, the 3Israeli death toll fluctuated in terms of severity and frequency. Terror was likely a leading cause of the Israeli economy’s recent worst economic performance, although these years also coincided with end of the high-tech bubble that also adversely affected the Israeli economy. In 2001 and . * I am grateful to Olivier Blanchard, John Hassler, Jerry Hausman, Arye Hillman, Guido Kuersteiner, Whitney Newey, and Joseph Zeira for their helpful advice and comments. I would also like to thank the participants in macroeconomics, econometrics, and finance workshops and seminars at MIT and Bar-Ilan University for useful observations. I would like to thank Simcha Bar-Eliezer from Israel CBS, Tomer Gardi from Btselem, Soli Peleg from Israel CBS, and Yael Shahar from ICT for providing data and helpful information. All remaining errors are mine. 1 This paper may include political terms. These are given only for the purpose of economic analysis and there is absolutely no political presumption in their usage. 2 Systematic risk will be taken to be synonymous with uncertainty regarding future events of relevance to consumers and investors. 3 See section for further details on the pattern of terror casualties in terms of number of dead and wounded and type of target. 2
200180160140120100806040200Total Dead In IsraelTotal Dead In WBGS Figure 1: Total Israeli dead in terror attacks against Israelis in Israel and West Bank and Gaza Strip Source: Database was compiled by the author from various sources listed in section . 2002 GDP decreased two years in a row for the first time since 1953, private consumption declined in 2002 for the first time since 1980, and the annual inflation for 2003 was negative for the first time since 1948. Section 2 introduces the theoretical background. Section 3 analyzes the pattern of terror in the Palestinian- Israeli conflict. I identify two types of terror attacks with different implications for perceived uncertainty with respect to personal security and future income. I then analyze the different effects of these uncertainties on consumption and investment. The analysis focuses on aggregated and disaggregated consumption data-- durables, nondurables, and services as well as fixed capital and inventories. Section 4 concludes. 31980 -11982 -11984 -11986 -11988 -11990 -11992 -11994 -11996 -11998 -12000 -12002 -1
2. Theoretical background Macroeconomic effects of terror have been studied by Abadie and Gardeazabal (2003), Eckstein and Tsiddon (2004), Eldor and Melinck (2004), Enders and Sandler (1996, 1991), Fielding (2003), and Fleischer and Buccola (2002). Eldor and Melinck (2004), Fielding (2003), and Fleischer and Buccola (2002) in particular also focus on terror against the population in Israel. These empirical studies have investigated the reduced form effect of terror on economic variables such as GDP, foreign investment, saving, financial markets, and tourism. In contrast, this paper considers the effect of terror on aggregate consumption and investment. Eckstein and Tsiddon (2004), which was written concurrently with and is closely related to this paper, shows that terror helps to explain changes in trend and business cycle of the Israeli economy. According to their paper, when terror peaks, the long-run equilibrium with an optimizing government is of lower output and welfare. Terror is expected to affect households’ decisions on consumption and firms’ decisions on investment through consequences for personal security and future GDP. The effect of terror on personal security is especially important when terror hits civilians in the course of their everyday lives in a random pattern, as has occurred in Israel. Terror affects GDP through two main mechanisms: increased uncertainty over future GDP growth (higher variance) and depressed expected GDP growth (lower expected mean). . Terror increases uncertainty Terrorists intimidate people by threatening their wealth and lives (no matter what is their motive and whether legitimate or not). They do so by causing casualties through attacks against civilian and military targets. In many cases, the population cannot ex-ante learn whether a change in the distribution of casualties has taken place; people only ex-post infer the change in uncertainty by observing terrorists acts, government responses, and outcomes including terror casualties. An 4
4increase in the number of casualties increases uncertainty that the population faces. This increase takes place through three channels: 1. Increasing contemporaneous uncertainty: There is a positive correlation between the number of terror casualties and the uncertainty that people face during the contemporaneous quarter. 2. Shifting the probability of future terror casualties upward (greater expected mean of terror casualties): The people’s posteriors are usually centered around the contemporaneous level of terror casualties. 3. Making the future less certain (increasing the variance of the distribution of future terror casualties): There is a positive correlation between the mean of current casualties and the variance of future casualties. There are two reasons for this observation. First, terror casualties are bounded from below by zero and unlimited from above. Second, the higher the current number of terror casualties is, the faster the terror and counter-terror may escalate. Terror may escalate due to temptation and higher morale on part of the terrorists, whereas counter-terror may escalate due to public support and lower tolerance for terror attacks on part of the government. Such escalations may increase, offset, or indeed decrease the terror threat in the short run. If terror casualties persist at the new high level without further escalation on part of the government or the terrorists, the state of terror will appear stable; hence the perceived uncertainty decreases. This uncertainty with respect to actual evolution of terror casualties, perhaps most importantly, generates in turn uncertainty over future personal security and GDP growth through the following three channels: 4 In some case, a sudden change of attack method is sufficient to bring about a new assessment of the uncertainty level, even if there was no change in terror casualties, ., the use of anthrax in the . in late 2001. As will be clear in Section 3 below, in the case of Israel, the number of casualties was a sufficient indicator for a possible change in uncertainty. 5
1. Economic instability: As shown in Abadie and Gardeazabal (2003), GDP is sensitive to the level of terror casualties. Since the path of terror casualties is uncertain, the path of the aggregate supply and demand is uncertain as well. Also, the impact on GDP is not immediate and lasts for some period over which the magnitude and timing of the impact is uncertain. 2. Political instability: A failure to counter terror or an unpopular move by the 5government will decrease its credibility and shock its stability. GDP growth rates, as 6is well documented, are negatively impacted by measures of political instability. 3. Personal insecurity: That terror decreases personal security is self-evident. This increase in insecurity prompts the consumers and investors to undertake measures to protect themselves, such as decreasing their use of public transportation, shopping in malls, eating out in restaurants, and perhaps attending weddings and other family celebrations. I shall address the macroeconomic effects of terror by analyzing the change in, rather than the level of, the number of casualties. The former calls on consumers and investors to alter their 7perceived uncertainty. The effect of uncertainty on durables consumption and irreversible investment is immediate and can be substantial, as will be evident from section . . Depressing GDP growth On the one hand, an increase in the number of terror casualties will depress GDP growth; the longer this unanticipated shock is expected to last, the stronger the effect will be on the GDP growth rate. On the other hand, such an increase might lead the government to increase resources 5Terror attacks in Israel over the past 15 years, for example, caused a change of a prime minister every two years on average, and no previous prime minister, except Ariel Sharon, was successful in a subsequent election. The wars on the regime of Saddam Hussein in Iraq and the subsequent terror were political issues in the . and the . 6See the literature on the effect of political conflict on economic variables, ., Acemoglu and Robinison (2001), Alesina et al. (1996), and Barro (1991). 7Abadie and Gardeazabal (2003), Enders and Sandler (1996, 1991), and Fleischer and Buccola (2002) considered only the effect of the level of casualties. 6
directed at preempting terror and reduce terror to a level lower than prior to the recent attack, in which case the growth rate of GDP will increase. The answer to which effect will dominate is, in general, unclear. However, whichever effect dominates, its impact will be immediate: according to rational expectations theory and the permanent income hypothesis, forward-looking agents immediately incorporate the anticipated change in GDP growth rate into their decisions. Such a change is expected to be positively correlated with durables and nondurables consumption as well as reversible and irreversible investment. While effects of a higher uncertainty and a lower GDP growth rate are expected to be immediate, the former affects mainly durables consumption and irreversible investment, while the latter affects all consumption and investment. This difference will prove helpful in distinguishing which effect is at play. . The impact of uncertainty . Temporary uncertainty Leahy and Zeira (2003), in a general equilibrium model that explicitly examines the timing of durables and nondurables consumption, observed that a temporary shock to wealth or income causes a bust-boom cycle in durable spending due to bunching of purchases in later periods. As consumers accumulate wealth, demand for durables returns to its pre-shock level. In the case of permanent shocks, there is a one-time reduction in durables consumption; there is no bunching and no boom. They also observe that wealth shocks have ambiguous effects on nondurables 8consumption. An alternative approach to examining the effects of uncertainty on durables is through irreversible investment theory. In the short-run, durables consumption, just like irreversible 8 When the price of durables relative to nondurables is assumed exogenous, changes in wealth cause a delay in the purchases of durables without changing their quantities, such changes have no effect on the quantities or timing of nondurables. This is what Leahy and Zeira called the perfect “insulation effect.” In the general case, when this assumption is relaxed, the effect on nondurables is ambiguous. 7
9investment, is characterized by durability, irreversibility, and indivisibility. Therefore, its response in the short-run to aggregate shocks is more in line with the aggregate implications of lumpy investment models than with the predictions of a frictionless life cycle-permanent income hypothesis (LC-PIH) such as Hall (1978) and Flavin (1981), especially in periods with high aggregate uncertainty (see Bar-Ilan and Blinder, 1988 and 1992; Bernanke, 1984 and 1985; Lam, 101989; Mankiw, 1982). Therefore, models that relate uncertainty to irreversible investment should be applicable in a straightforward manner to the relationship between uncertainty and consumption. Bernanke (1983) and Cukierman (1980) have demonstrated the option value of deferring irreversible investment when there is temporary uncertainty and when information pertinent to the choice of investment arrives over time and makes the future less uncertain. They showed that a temporary increase in uncertainty could cause a bust-boom cycle in irreversible investment spending. In an application of this model to durables consumption, we should expect a similar pattern of response by consumers to temporary increases in uncertainty due to increases in terror victims. . Permanent uncertainty There are two principal different, but not mutually exclusive, models relating a permanent increase in uncertainty with irreversible investment, which can be applied to examine the relation of a permanent increase in uncertainty with durables consumption: the value of waiting and the (S,s) models. A model that directly examines the effect of a permanent increase in uncertainty on consumption is the precautionary saving model, as explained below. 9 Durability separates consumption from purchase, irreversibility means long-term commitment, and indivisibility implies lumpy and discontinuous purchases. 10 Note that Caballero (1990) validates the LC-PIH for the medium-long-run. Several justifications for the slow adjustment of durables consumption have been advanced in the literature, including inter-temporal complementarities in consumption, convex adjustment costs, and financial markets imperfection. Surveys of the modern consumption literature by Deaton (1992) and Hall (1989) discuss some of the efforts that have been made to explore departures from the frictionless LC-PIH. 8
Pindyck (1991) developed a model related to Bernanke (1983) and Cukierman (1980), in which information arrives over time but the future is always uncertain. Here too there is a value associated with waiting and a permanent increase in uncertainty of the economic environment has a negative long-term effect on the amount of actual irreversible investment, which should be symmetrically applicable to durables consumption. Bar-Ilan and Blinder (1988, 1992), Bertola and Caballero (1990), Caballero (1993), and Eberly (1994) applied the (S,s) model to durables purchases. They concluded that higher risk broadens the (S,s) bands, so durables deviate more from their optimal levels before being adjusted. Hence, a permanent increase in uncertainty causes a drop in fixed investment as well as durables 11consumption. The precautionary saving model directly examines the effect of permanent increases in uncertainty on consumption. Under inter-temporal optimization and decreasing CARA utility function, an increase in uncertainty leads consumers to defer consumption (to be more prudent and increase precautionary saving). Put differently, when risk increases, consumption declines, after which its growth rate increases without a bust-boom cycle in the short-run. This result is reinforced if it is also assumed that consumers have adaptive expectations, which is likely when 12consumers consider the state of terror. If changes in subjective uncertainty fall more heavily on some individuals, as is likely the case in terror attacks (for example, children and workers using public transportation), their aggregate effect can be much larger (Blanchard and Mankiw, 1988). Therefore, if changes in terror casualties are interpreted as a permanent increase in uncertainty, we should expect consumption to decline in the short run before it adjusts to its long-term level. Thus, a permanent increase in uncertainty causes a decline in irreversible investment and durables consumption in all three models: the value of waiting, the (S,s), and the precautionary 13saving models. However, there is an important difference in implications between the value of 11 The (S,s) type of models generally considers only permanent shocks. 12 In such a case, consumers expect further decline in their income, which enhances their precautionary saving. 13 The depressing effect of uncertainty is well-documented in the theoretical literature. Ingersoll and Ross (1988) have examined irreversible investment decisions when the interest rate evolves 9
waiting and the precautionary saving models: the former is relevant only for durables consumption and irreversible investments, whereas the latter is relevant to nondurables and inventories as well as durables consumption and fixed capital. This difference in reaction to uncertainty will be useful when empirically distinguishing irreversibility effects from precautionary saving effects. In all of the above models, the effect of macro-uncertainty, whether temporary or permanent, on aggregate irreversible spending follows from the micro-analysis. Terror attacks are a classic case where uncertainty about future macro-information pertinent to the choice of durables consumption and irreversible investment is not eliminated by aggregation level and the law of 14large numbers does not apply. 3. Empirical findings Due to the lack of data availability that relates macro-uncertainty with durables consumption and irreversible investments, little empirical evidence is provided in the literature for the effect of macro-uncertainty on durables consumption and irreversible investment. A notable exception is Romer (1990). She suggested, based on Bernanke (1983), that the fall in durables consumption in the year following the October 1929 Great Crash could be best explained by the temporary 15increase in uncertainty regarding the course of future income. That the Great Crash caused uncertainty was evident from the decline in surety expressed by contemporary forecasters. stochastically, but future cash flows are known with certainty. As with uncertainty over future cash flows, the uncertainty regarding future interest rate creates a value to waiting. Investing is depressed further as the volatility of interest rates grows. Caballero and Corbo (1988) have shown how uncertainty over future real exchange rates can depress exports. Dornbusch (1987) has noted that uncertainty over future tariffs structure creates an opportunity cost to committing capital to new physical plants. 14 For examples on idiosyncratic shocks that eliminate cycles, see the (S,s) models in Caballero (1993), Caballero and Engle (1991), Caplin and Spulber (1987), and Zeira (1990). 15 Romer concluded, based on informal argument, that Bernanke’s (1983) model implies that uncertainty is positively correlated with nondurables consumption. This conclusion seems erroneous, as illustrated above. Further, Bernanke (1985) found that durables and nondurables are neither strong substitutes nor strong complements; thus the ‘spillover’ effect from slowly adjusting durables to nondurables may not be especially important in practice. In fact, Romer, too, found that uncertainty had had an ambiguous effect on nondurables consumption, a finding that is supportive of our conclusion above. 10
Romer’s case study included only one event and she did not detect a bust-boom cycle in 16consumption. Here I examine the effect of changes in macro-uncertainty on durables consumption over 23 years, a period over which macro-uncertainty fluctuated due to changes in terror fatalities. . Data description For the purposes of this study, I constructed a new daily database that provides an accurate and complete coverage of politically-motivated terror attacks carried out by Palestinians against Israeli targets during the years 1980-2002. This database is based on the following websites: ABC News; ADL; Al-Jazeera (Arabic newspaper); B’Tselem (The Israeli Information Center for Human Rights in the Occupied Territories); BBC News World Edition; CNN; GPO - Government Press Office; Haaretz (Israeli Hebrew newspaper); HAMAS (Islamic Resistance Movement); ICT - The International Policy Institute for Counter-Terror; IDF - Israel Defense Forces; Islam Online; Jerusalem Post (Israeli English Newspaper); LAW (The Palestinian Society for the Protection of Human Rights and the Environment); MFA - Ministry of Foreign Affairs; Peace Now; PIC - Palestinian Information Center; Shia News; The Ministry of Labor and Social Affairs and the National Insurance Institute; TVA - Terror Victims Association; Walk for Israel; and others. The data includes 687 observations. Each observation includes the following information: date, town, number killed, number injured, method of attack, type of target, and the particular Palestinian organization carrying the attack. The details (and most importantly, the number of fatalities) of each observation in the data was confirmed by at least two sources with the exception of those observations recorded by B’Tselem. The reason for this exception is that B’TSelem, an independent human rights organization, ensures the reliability of information it publishes by conducting its own fieldwork and research, whose results are thoroughly cross- 16 Hassler (2001), following Romer (1990), relates automobile purchases to uncertainty that can be learned from fluctuations in the exchange. 11
checked with relevant documents, official government sources, and information from other sources, among them Israeli, Palestinian, and other human rights organizations. For all observations, except five, the number of reported fatalities was identical for all sources. For those five observations, I use the number of fatalities that was recorded by more sources (the discrepancy was less than 2 fatalities). In some cases, the body of kidnapped soldier was found in a quarter after the quarter during which the kidnapping happened, or some of the casualties died later of their wounds and their day of death fell in a quarter following the quarter during which the attack occurred. In such cases, the day of the kidnapping or attack was used. I maintained a distinction between Israel within its 1948 borders and West Bank and Gaza Strip (WBGS) that was occupied by Israel in 1967. I also distinguish between Jewish and Arab neighborhoods of Jerusalem. A terror attack on Jews that took place in East Jerusalem was classified as “in Israel” if the attack occurred in a Jewish neighborhood, and “in the West Bank” if Jews were attacked in an Arab neighborhood. 17Data on national accounts were provided by the Israeli Central Bureau of Statistics. All national accounts data employed were chained at 2000 prices. . The pattern of terror Terror attacks against the population of Israel have had ebbs and flows in terms of severity and frequency (see Figure 1). Between January 1980 and December 2002, terror left a total of 914 dead and 4,755 wounded in attacks against civilians, in addition to 294 dead and 515 wounded in attacks against military personnel (see further details in Table 1). Terror attacks that caused yet again a renewed peak in monthly fatalities called for Israeli households to alter their perceived (subjective) level of personal risk and the possible effect on their future income. Note that it is also likely that terror attacks carried out by Jews against civilian and military Palestinians targets 17 These are the most recently revised data series. National accounts data for 1980-1995 were compiled according to SNA68, whereas those for 1995 and onwards are based on SNA93. 12
have caused Israeli households update their beliefs, as Israeli households expected revenge. However, appropriate data on such terror attacks was not available. Table 1: Total Number of Violent Attacks, Dead, and Wounded by Target of Attack 1980-2002 ForcesMilitary Personnel 2 19 286 505 4 0 47 110 179 239 395Police Personnel/Facility 1 2 8 10 4 3 4 8 5 Personnel 2 2 1 1 1 - 1 1 in GeneralCivilians* 2 30 264 616 100 1 07 431 130 157 TransportationBus 4 4 246 1,018 2 8 212 934 1 6 34 84Bus stop 1 9 61 535 1 9 61 535------Train Station 1 3 90 1 3 ZoneShopping Center 1 4 53 986 1 3 52 986 1 1 -Marketplace 1 1 29 423 9 28 423 2 1 Facility 5 42 236 5 42 236------Restaurant 1 0 49 379 8 48 375 2 1 4Hotel 2 29 160 1 29 150 1 - 1 0Beach 3 1 - 3 1 6 9 83 97 8 11 48 61 72 49Cargo Transport 7 6 - 2 - - 5 6 ZoneIndustrial Zone 5 1 13 2 - 9 3 1 1 - 1 1 1 - 1 of Worship 1 11 50 1 11 50------Pupil 1 2 4 1 2 4------School 1 - 8 1 - 8 ------University 1 9 86 1 9 86------Borders 6 23 42 5 20 42 1 3 -Grand Total 6 64 1,208 5,270 254 6 87 4,532 4 10 521 738* Including infiltrations, shootings, stabbings, and other events in different places. Source: The database was compiled by the author from various sources listed in section . 13
Most of the attacks in Israel took place in Jerusalem and in a 10-mile-wide coastal strip that stretches along a 100 miles length from Haifa in the north to Ashdod in the south. During the period under investigation, over 80% of the Israelis, and only Israelis, lived or worked daily in this area. As Figure 2 18shows, terror attacks were primarily on civilian targets (accounting for more than 90% of the death 19toll). The most lethal attacks targeted public transportation, commercial zones and entertainment facilities. Therefore, attacks inside Israel had direct implications on most Israeli households’ personal 20security and future income. Attacks against military or production targets in Israel numbered only a few (accounting for about 7% of dead). TouristOthersArmed Forces0%6%7%Industrial ZoneGovernment0%0%Vehicle2%Civilian in GeneralEntertainment16%17%Commercial Zone12%Public Transportation40% Figure 2: Total terror fatalities in Israel by target of attack, 1980-2002 18 The distribution of wounded provides basically a similar pattern. 19 Different methods were used, of which shooting and grenades, suicide-carried explosives, and knife attacks were the most common as well the most deadly, accounting for more than 80% of all categories: number of attacks, number of dead, and number of wounded. 20 One might argue that fluctuations in uncertainty fell more heavily on some households, while others (., those who could afford to work in the countryside and use their private cars) were able to avoid it. If such were the case, then the impact of fluctuations in aggregate uncertainty on consumption was aggravated (see Blanchard and Mankiw, 1988). 14
Conversely, less than 3% of Israelis lived in settlements in the areas of the WBGS. A total population of approximately 200,000 lived in 150 Israeli settlements spread among nearly 700 Palestinian towns with a total population of just over three million Palestinians, as of March 21, 222000. Both Israeli civilian and Israeli military targets were attacked in WBGS and equally suffered in the death toll (see Figure 3). ToOthersurist1%0%Industrial ZoneVehicle0%15%Entertainment0%Commercial ZoneArmed Forces0%47%Public Transportation7%Civilians in General30%Government0% Figure 3: Total Terror Fatalities in WBGS by Target of Attack, 1980-2002 Notably, only the demand side (for consumption of durables, nondurables, and services) of the Israeli economy was targeted; industry, tourists, politicians, diplomats, and international interests 21 Settling WBGS by Israeli settlers started in 1967. Israel has recently withdrew from Gaza Strip but continue to settle and occupy the West Bank. In addition to the above official settlements, there have always been tens of unofficial outposts and settlements, each with only a few settlers. 22 These figures are based on data obtained from the Israeli Central Bureau of Statistics, the Palestinian Bureau of Statistics, and Peace Now Settlement Watch research. 15
were rarely targeted. The terror attacks therefore had a direct and first-order effect on the demand side of the economy. Furthermore, due to the above demographic and attack patterns, it is expected that Israeli household consumption demand would respond more substantially to news about fatal attacks in Israel than in WBGS; the former meant a more imminent increase in perceived personal risk and possible retaliation. Israeli households and investors perceived terror to be temporary and they expected the Israeli government to constantly improve its effectiveness in counter-terror. Figure 1 supports these perceived beliefs, as it shows that there were periods when terror escalated, and, later on, were followed by de-escalation and calm periods. Theory relating uncertainty to consumption, in conjunction with the economic consequences of the terror inflicted on the Israeli population, underlies the following four null hypotheses: 1. Null hypothesis 1: An increase in the number of terror fatalities has a negative impact on total consumption. 2. Null hypothesis 2: Durables are more affected by terror fatalities than nondurables. 3. Null hypothesis 3: Terror fatalities in Israel have a higher impact than those in WBGS. 4. Null hypothesis 4: A temporary increase in terror fatalities causes a bust-boom cycle in consumption, whereas a permanent increase in terror fatalities causes a one-time drop in consumption. The next section will examine these null hypotheses. Before proceeding to the empirical results, two further points are worth noting. First, since the economy of the perpetrators of the terror had only marginal feedback on the Israeli economy, temporal variations in consumption of Israelis can be expected to have had little effect on the intensity of fatal attacks; therefore, endogeneity is unlikely to be a serious issue on a quarterly basis. Second, the significant increase in terror fatalities that began in the year 2000 coincided with the end of the Hi-Tech boom. The peak in terror fatalities and the end of the Hi-Tech boom are independent exogenous influences to the economy that are known not to be uncorrelated. Furthermore, while the end of Hi-Tech occurred through a short period that lasted 16
one or two quarters without further fluctuations, terror fatalities after the year 2000 continued to fluctuate for several quarters. Therefore, as noted by Eckstein and Tsiddon (2004), controlling for fluctuations in the Hi-Tech may reduce the negative values of the coefficients of terror 23fatalities and the marginal significance levels but would not reverse the basic results. . The effect of terror on consumption This section analyzes the impact of terror attacks on household consumption at the aggregate and disaggregated levels (see Figure 1). We first test and reject the null hypothesis that terror fatalities series is non-stationary. To test the above four hypotheses, we use quarterly data to regress the logarithmic difference of the real consumption (growth rate) on the logarithmic 24change in terror fatalities. To avoid a logarithm of zero in some quarters, 1 is added to the terror fatalities series. To test for possible lag responses due to durability or market imperfection, we include contemporaneous and lagged regressors (until a lag coefficient is insignificant at 5% level). Possible auto-regression and moving average (ARMA) components are examined according to Box-Jenkins (1976) methodology with 5% significance level. We find in all cases that MA(1) or/and MA(2) should be included. The Newey-West (1987) estimator, which is robust for the remaining heteroskedasticity and autocorrelation of any type, is used and reported. Quarterly dummies are also included to de-seasonalize the data. 23 Eckstein and Tsiddon (2004) control for fluctuation in the Hi-Tech sector by including the first differences in logs of the NASDAQ index in real dollar terms for the period 1980:1 to 2003:3 as an additional exogenous variable in the VAR system with two and one lags. 24 Another possibility is to regress the level of consumption on a time trend and the level of fatalities. On a macroeconomic theoretical basis, this should not change the results, and indeed it does not provide qualitatively different results. However, on an econometric theoretical basis, using the first differences as the dependent variables in a time series regression is preferable, because of their better treatment of serial correlation. 17
Null hypothesis 1: An increase in terror fatalities has a negative impact on total consumption We test the first null hypothesis that terror fatalities have a negative impact on total consumption, 25with possible lagged response, by estimating the following regression: 43KdLn(C)=α+β*dLn(TF)+χQd+ε+δε(1) t∑it−i∑dtt∑kt−ki=0d=1k=1where C is total consumption of Israelis, TF is the total number of Israeli terror fatalities in both ttIsrael and WBGS, Qd is a quarterly dummy variable that equals 1 for quarter d and zero totherwise, and δ is the moving average component. The results of this regression are shown in kcolumn 1 in Table 2; for brevity, I only report the estimates of the constant and the coefficients of dLn(TF)'s. The contemporaneous through the last significant lagged estimated coefficients of t-idLn(TF) are negative. Also, the one-year elasticity (. long-term effect), which is the sum of t-iall five coefficients of dLn(TF), is negative and significant. Therefore, we reject the null that t-ifatal terror has no negative impact on total consumption. This impact starts after a delay of one quarter and lasts for not longer than a year, as only the first through the third lagged estimated coefficients are significant. The estimated one-year impact of terror on consumption is substantial, as the one-year elasticity is about –% (see last row, first column, in Table 2), which means that a permanent 100% increase in quarterly fatalities (for instance, from 10 to 20) will bring about % decrease in total private consumption in the year following the increase in fatalities. Applying this estimate to the actual fatality numbers during 2000/Q1-2002/Q4 of El-Aqsa Intifada, the Palestinian uprising against the Israeli occupation, reveals a total contribution of roughly –9% to consumption growth rate. This finding, along with the fact that the average annual consumption growth rate was about 4% per annum for the five years preceding the El-Aqsa Intifada, explains the slightly negative (less than -1%) consumption growth rate during 2000/Q1-2002/Q4 of El-Aqsa Intifada, for the first time in two decades. 25 Note that an error correction model is not applicable in this case, since consumption is integrated of order one, ., I(1), while terror fatalities is stationary, ., I(0). 18
Table 2: The impact of terror fatalities on total consumption dLn(Total consumption) t 1980/Q1- 2002/Q4 1980/Q1-2002/Q4 1980/Q1-2002/Q4 1987/Q1-2002/Q4 GARCH(1,1)with IV for GDP (1) (2) (3) (4) C * * * * () () () () dLn(TF) t() () () () dLn(TF) * ** ** * t-1 () () () () dLn(TF) * ** * ** t-2() () () () dLn(TF) * * ** ** t-3 () () () () dLn(TF) t-4() () () () 2R D-W #Obs. 87 87 85 66 One-year Elasticity * * * * () () () () Notes: One-year elasticity is the sum of all five coefficients of dLn(TF). Newey-West HAC Standard Errors & t-1Covariance are used (lag truncation=3). P-values are in parenthesis; * and ** indicate variables significant at 5% and 10% significance level, respectively. To avoid a logarithm of zero in some quarters, 1 is added to TF series. All data employed were chained at 2000 prices from original series. Quarterly dummy de-seasonalizing variables and MA components are included but not reported for brevity. The instrumental variables in column (3) include 2-4 lagged GDP growth rates and 2-4 lagged consumption growth rates. Different tests are used to examine the robustness of the above findings. The tests include GARCH specification and controlling for business cycles (by including contemporaneous 19
26GDP). The results of these tests, presented in columns (2)-(3) of Table 2, show no notable changes in the substance or significance of the coefficients. Further, I test for a possible breakpoint after the first Intifada (uprising) in 1987. T test is performed in two ways. First, regression (1) is run above for the period 1987-2002 rather than 1980-2002, yielding no notable changes in the results (compare results in column (4) with column (1) of Table 2). Second, the following regression is run for the entire period 1980-2002: 44dLn(C)=α+β80*(1−D87)*dLn(TF)+β87*D87*dLn(TF)∑∑titt−iitt−ii=0i=0(2) 3K +χQd+ε+δε∑d∑ttkt−kd=1k=1where D87 is a dummy variable equals zero for the period 1980-1986, and one for the period t441987-2002. Next, we test the null hypothesis: H:β80=β87, which is not rejected at 0∑i∑ii=0i=05% percent significance level. Other tests were also performed but are not reported in Table 2. To test the null hypothesis that consumption growth is affected by the level of, rather than the change in, terror fatalities, the fifth lag of terror fatalities (TF-) was added as an additional regressor. If the coefficient of this t5regressor is significant, we cannot reject the null hypothesis that consumption growth rate should be regressed on contemporaneous and one-through-five lagged TF in levels; that is, we have co-integration. The results show that this additional regressor is insignificant for all cases, rejecting 27the null hypothesis. 26 When GDP was included, I used instrumental variables estimation. The instrumental variables include 2, 3, and 4 lagged GDP growth rates and 2, 3, and 4 lagged consumption growth rates. We also run the same regression by using consumption-GDP ratio as dependent variable and find no change in the results. 27 We also directly examine the above null by regressing the consumption growth rate on the level of terror fatalities (contemporaneous and five lags), and find that all coefficients of terror fatalities are insignificant, which rejects the null that a temporary terror attacks impact the long-run (steady-state) consumption level or growth rate. 20
To test the importance of the frequency of attacks, we include the number of attacks in regression (1) above and find it insignificant at 10% level. We also include quadratic and cubic values of TF‘s and find them insignificant at 10% level. t-iNull hypothesis 2: Durables are more affected by terror fatalities than nondurables. Next, we proceed to test our second hypothesis: the higher the durability of consumption, the greater the impact of terror fatalities is. To do so, we regress the following three regressions: 43KdLn(Dur)=α+β*dLn(TF)+χQd+ε+δε(3) ∑∑∑tit−idttkt−ki=0d=1k=143KdLn(NonDur)=α+β*dLn(TF)+χQd+ε+δε(4) ∑tit−∑∑idttkt−ki=0d=1k=143KdLn(Ser)=α+β*dLn(TF)+χQd+ε+δε(5) t∑it−i∑dtt∑kt−ki=0d=1k=1where the variables Dur, NonDur, and Ser are durables consumption, nondurables tttconsumption, and services consumption, respectively, in the domestic market. The results of these three regressions are presented in Table 3. The results of the regressions are consistent with the theory. First, they provide similar qualitative results as those of regression (1); namely, terror fatalities negatively affect consumption of durables, nondurables, and services. This is evident from the finding that the contemporaneous through the last significant coefficients of TF as well as the long-term effects, measured by the one-year elasticities, are negative in all three regressions. Second, durables consumption seems to adjust slower than nondurables consumption. This is because the coefficient of one, two, and three lagged terror fatalities are significant in the durables regression, whereas it is only the coefficient of one lagged terror fatalities that is significant in the nondurables regression. 21
Table 3: The impact of terror fatalities on durables, nondurables, and services, 1980/Q1-2002/Q4 dLn(Durables dLn(Nondurables dLn(Services consumption) consumption) consumption) ttt(1) (2) (3) C * * ()()()dLn(TF) ()()()dLn(TF) * * * ()()() dLn(TF) * ** ()()() dLn(TF) ** * ()()() dLn(TF) ()()() 2R D-W #Obs. 878787 One-year Elasticity * * * ()()() See notes on Table 2. Third, the strongest long-term impact, measured by the one-year elasticity, is on durables consumption (), followed by nondurables consumption () and services consumption (). In other words, a 100% increase in the number of fatalities (., from 10 to 20 fatalities per quarter) will cause a drop by 8%, %, and % in durables, nondurables, and services consumption over the proceeding year, respectively. The robustness of the above findings is examined by testing the impact of fatal attacks on the decomposition of durables consumption: furniture, household equipment, and personal 22
28transportation equipment. We find that the qualitative results remain unchanged; namely, there is a negative impact on all of these three sub-categories with at least one-quarter delay. An examination of the sub-categories of the nondurables series shows that the food, beverages, and tobacco category (the least durable sub-category of the nondurables consumption) is hardly affected. Although services consumption was the least affected as predicted by the theory, it is still interesting to understand why it is at all affected and the channel through which it is affected. To examine this matter, we consider the composition of the data series of services and find that services fluctuate mainly due to fluctuations in dining and accommodations services, which is substantially affected by foreign demand in the domestic market. Fleischer and Buccola (2002) found that foreign demand for accommodation at Israeli hotels is sensitive to terrorist activity, whereas domestic demand is insensitive. They also report that domestic demand provides only little buffer for declines in foreign tourism. Therefore, we expect that the elasticity of foreign consumption in the domestic market with respect to terror fatalities is higher than the elasticity of dining and accommodations services with respect to terror. I estimate both elasticities and present the results in Table 4. As expected, both elasticities are significantly negative and the elasticity of foreign consumption is higher than the elasticity of dining and accommodations services. Therefore, it appears that the impact of terror fatalities on services is primarily through the effect on foreign demand in the domestic market. 28 For brevity, the results are not reported and can be obtained from the author. 23
Table 4: The impact of terror fatalities on consumption- sub-categories, 1980/Q1- 2002/Q4 dLn(Non-profit dLn(Foreigners’ dLn(Dining and institutions consumption in the accommodation serving domestic market) services) household) 1980/Q1-2002/Q4 1995/Q1-2002/Q4 1980/Q1-2002/Q4 (1) (2) (3) C * * * ()()()dLn(TF) * * ()()()dLn(TF) * ()()()dLn(TF) * ()()() dLn(TF) * ()()() dLn(TF) t-4 ()() 2R D-W #Obs. 88 30 87 One-year Elasticity * * * ()()() See notes on Table 2. Note also that, unlike Israelis who respond in delay, foreign consumers adjust immediately. The coefficient of the contemporaneous terror fatalities is significantly negative in both regressions for foreign consumption in the domestic market as well as for dining and accommodation services. Further, not only is the timing of response different, but also the magnitude of the response: foreigners’ elasticity () is about five times that of Israelis (). Therefore, we should expect an increase in terror activity to have an immediate and intense negative effect on 24
the outputs of all industries for which there is foreign demand, such as airlines, hotel 29accommodation, and restaurants. Finally, we examine the impact of fatal attacks on the consumption of non-profit institutions serving households. We find that it is the only category of consumption that is positively impacted by fatal terror attacks, as its one-year elasticity is (see Table 4, column 3). The rationale behind this exceptional positive sign is clear: when people are hit by terror attacks, more non-profit institutions will come to their aid. Null hypothesis 3: Terror fatalities in Israel impact consumption more than those in WBGS As noted above, perceived personal insecurity and uncertainty about future income is more sensitive to fatal attacks in Israel than in WBGS. Therefore, we expect that the impact of fatal terror attacks in Israel to be stronger than the impact of fatal attacks in WBGS. We test this null hypothesis by running regressions (3)-(5) using two alternative series: Israeli fatalities in Israel and Israeli fatalities in WBGS, instead of the sum of both. Since the impact of attacks against civilian targets may differ from the impact of attacks against military targets, and because attacks in WBGS are more military-target intense than attacks in Israel, we compare the results of regressions that employ fatalities in attacks against civilian targets in WBGS as against those of regressions that employ fatalities in attacks against civilian targets in Israel. The estimates of these regressions are presented in Table 5 below. 29 These results are not confined to the case when terror attacks involve the use of airplanes, as such was the case on September 11. 25
Table 5: The effect of terror fatalities on consumption by place of attack, 1980/Q1- 2002/Q4 Attacks against civilians in Israel Attacks against civilians in WBGS dLn(Durables dLn(Nondurables dLn(Services dLn(Durables dLn(Nondurables dLn(Services consumption) consumption) consumption) consumption) consumption) consumption) tttttt(1) (2) (3) (4) (5) (6) C * * * * ()()()()()()dLn(TF) * t ()()()()()()dLn(TF) * * * * t-1 ()()()()()()dLn(TF) * * * * t-2 ()()()()()()dLn(TF) * t-3 ()()()()()()dLn(TF) t-4 ()()()()()()2R D-W #Obs. 878787 878787One-year Elasticity * * * * * * ()()()()()()See notes on Table 2. 26
As expected, all significant coefficients and all one-year elasticities of consumption with respect to fatalities are negative, whether using fatalities in terror attacks against civilian targets in Israel or in WBGS, and whether the dependent variable is durables, nondurables, or services consumption. The one-year elasticities of durables and nondurables with respect to fatalities in attacks against civilian targets in Israel ( and ) are not significantly greater than their counterpart elasticities with respect to fatalities in attacks against civilian targets in WBGS ( and –). However, the one-year elasticity of services is equal for both fatalities in Israel and WBGS. A possible explanation for this equality is that foreigners, by whom this elasticity is most determined, do not distinguish between terror attacks that take place in Israel from those that take place in WBGS. Null hypothesis 4: A temporary increase in terror fatalities causes a bust-boom cycle in durables consumption, whereas a permanent increase causes a one time drop We simulate an impact on durables consumption in two scenarios: a temporary increase and a permanent increase in terror fatalities. We use the coefficients on Table 3 to report the results of these simulations in Figure 4 and Figure 5. The construction of our regression produces results that are consistent with the theory. The graphs of durables consumption show a bust-boom cycle and a one-time drop after, respectively, a temporary and a permanent increase in terror fatalities. Finally, it is possible that there is asymmetry in the response of durables consumption to changes in terror fatalities: the impact of an increase is different than the impact of a decrease in terror fatalities. In Appendix A, we run simulations that allow asymmetry. These simulations reveal a similar pattern for the durables consumption as in Figure 4 and Figure 5, with two exceptions. First, the bust-boom cycle is stronger. Second, a temporary increase in terror causes durables consumption to retreat to a lower than the initial level, which suggests a stronger income shock effect (see Figure 6). 27
200Durables consumption with180stable violence fatalities160140120Durables consumptionwith a temporary 100increase in Violence Fatalities20--3-2-1012345678 Figure 4: Simulation of a temporary increase in terror fatalities 200Durables consumption with180a stable violence160140Durables consumption120with a permanent increase in violence1008060Total Violence Fatalities 4020--3-2-1012345678 Figure 5: Simulation of a permanent increase in terror fatalities 28
. The effect of terror on investment This section examines how terror affects investment. To this end, the following regression is estimated: 43KdLn(FC)=α+β*dLn(TF)+χQd+ε+δε(6) t∑it−i∑dtt∑kt−ki=0d=1k=1where FC is fixed capital and the regressors are the same as in regression (1). Also tincluded are dummy variables to control for the 1991 Gulf war in regression (6). The results are presented in Table 6. All the coefficients of terror fatalities as well as the one-year elasticity in regression (6) are negative. Lags one through three and the one-year elasticity are significant. This suggests that, as predicted by the irreversible investment theory discussed above, terror fatalities negatively impact irreversible investment through the effect on uncertainty. The one-year elasticity of fixed capital investment with respect to terror fatalities () is higher than the one-year elasticity of total consumption with respect to terror fatalities (), which means that firms’ decision on investment is more sensitive than households’ decision on consumption. 29
Table 6: The impact of terror fatalities on investment, 1980/Q1-2002/Q4 dLn(Fixed capital) 1980-2002 (1) * ()dLn(TF) t ()dLn(TF) * t-1 () dLn(TF) * t-2() dLn(TF) * t-3 () dLn(TF) t-4() 2R -W #Obs. 87 One-year Elasticity () See notes on Table 2. 4. Summary and conclusions This paper has demonstrated quite robust evidence for the effect of terror on consumption and investment. Uncertainty due to terror significantly affects consumption and investment in a manner consistent with the theory that relates uncertainty and consumption. A temporary increase in terror fatalities, while having no impact on the long-run level and growth rate of consumption, causes a bust-boom cycle in consumption in the short-run. Durables consumption is most affected, followed by 30
nondurables. Services are affected primarily through the tourism channel, which affects dining and accommodation services. The responses of all series lag one quarter (except dining and accommodation services) and last no longer than three quarters. A permanent increase in terror fatalities causes a one-time drop in consumption. Fixed capital also exhibits a bust-boom cycle after a temporary increase in terror fatalities and a one-time drop after a permanent increase in terror fatalities. However, the change in inventories is positively impacted by terror fatalities, which is an indication that the demand side reacts faster than the supply side. 31
Appendix A: Asymmetric response by consumption In section 3 above, we assumed that the impact of an increase in terror fatalities has the same magnitude as the impact of a decrease in terror fatalities on durables consumption. Put another way, we assumed equal coefficients for both- positive and negative values of TF. We would like to test this hypothesis, while restricting ourselves to the case of durables. To this end, we run the following regression: 34PNdLn(Dur)=α+β*P*dLn(TF)+β*(1−P)dLn(TF)t∑it−it−i∑it−it−ii=0i=0(7) 3K +χQd+ε+δε∑dtt∑kt−kd=1k=1where P is one if there is an increase in terror fatalities at time t-i, and zero if there is a t-idecrease in terror fatalities at time t-i. All other variables are defined as before. The 1estimates of this regression are presented in Table 7. As it is clear from Table 7, will still maintain the basic results that there is a negative correlation between terror fatalities and durables consumption and that the impact starts after one quarter and lasts no more than four quarters. Further, the elasticities of both- an increase and a decrease in terror fatalities- are significantly negative. Furthermore, we reject the null that both elasticities are equal at 5% significance level. Simulations of the results of the above unrestricted regression for the case of temporary increase in terror fatalities are presented in Figure 6. The response of durables is similar to that in Figure 5 with two exceptions. First, the bust-boom cycle is stronger. Second, a temporary increase in terror causes durables consumption to retrieve to a lower level than the initial, which suggests a stronger income shock effect. 31 The number of lagged dependent variables was chosen such that until a lag coefficient is insignificant at 5% level and according to Box-Jenkins (1976) methodology. 32
Table 7: The impact of terror fatalities on durables consumption, assuming a different impact for a decease and an increase in terror fatalities, 1987/Q1-2002/Q4 dLn(Durables Consumption) (1) *dLn(TF) *dLn(TF) * Pβ=− ∑ *dLn(TF) 0 *dLn(TF) * (1-P)*dLn(TF) (1-P)*dLn(TF) (1-P)*dLn(TF) * t-2t-2Nβ=−003 .∑i (1-P)*dLn(TF) (1-P)*dLn(TF) * t-4t-4 2 R D-W #Obs. 64 See notes on Table 2. P is one if there is an increase in terror fatalities at time t-i, and zero if t-ithere is a decrease in terror fatalities at time t-i. 33
200Durables consumption with180a stable violence fatalities160140120Durables consumptionwith a temporary increase100in violence fatalities806040Total Violence Fatalities20--3-2-1012345678 Figure 6: Simulation of temporary increase in terror fatalities with asymmetric response by durables consumption 34
Bibliography Abadie, A., Gardeazabal, J., 2003. The economic costs of conflict: a case study of the Basque country. American Economic Review 93, 113-132. Acemoglu, D., Robinison, ., 2001. A theory of political transition. American Economic Review 91, 938-963. Alesina, A., Ozler, S., Roubini, N., Swagel, P., 1996. Political instability and economic growth. Journal of Economic Growth 1, 189-111. Bar-Ilan, A., Blinder, ., 1988. The life cycle permanent-income model and consumer durables. Annales d'Economie et de Statistique, 71-91. Bar-Ilan, A., Blinder, Alan S., 1992. Consumers durables: evidence on the optimality of doing nothing. Journal of Money, credit and Banking 24, 258-272. Barro, ., 1991. Economic growth in a cross section of countries. Quarterly Journal of Economics 106, 659-684. Bernanke, ., 1983. Irreversibility, uncertainty, and cyclical investment. Quarterly Journal of Economics 98, 85-106. Bernanke, ., 1984. Permanent income, liquidity, and expenditure on automobiles: evidence from panel data. Quarterly Journal of Economics 99, 587-614. Bernanke, ., 1985. Adjustment costs, durables, and aggregate consumption. Journal of Monetary Economics 15, 41-68. Bertola, G., Cabellaro, ., 1990. Kinked adjustment costs and aggregate demand. In: Olivier and Stanley (Eds.), NBER Macroeconomics Annual 1990. MIT Press, Cambridge, pp. 237-288. Blanchard, ., Fischer, S., 1989. Lectures on Macroeconomics, MIT Press, Cambridge MA. Blanchard, ., Mankiw, ., 1988. Consumption: beyond certain equivalence. American Economic Review Papers and Proceedings 78, 173-177. Box, G., Jenkins, G., 1976. Time Series Analysis: Forecasting and Control. Holden-Day, San Francisco. Caballero, ., 1990a. Consumption puzzles and precautionary savings. Journal of Monetary Economics 25, 113-136. Caballero, ., 1990b. Expenditure on durable goods: a case for slow adjustment. Quarterly Journal of Economics 105, 727-743. 35
Caballero, ., 1991a. On the sign of investment-uncertainty relation. American Economic Review 81, 279-288. Caballero, ., 1991b. Earnings uncertainty and aggregate wealth accumulation. American Economic Review 81, 859-871. Caballero, ., 1993. Durable goods: an explanation for their slow adjustment. Journal of Political Economy 101, 351-384. Caballero, ., Corbo, V., 1988. Real exchange rate uncertainty and exports: multi-country empirical evidence. Working Paper 414, Department of Economics, Columbia University. Caballero, ., Engle, E., 1991. Dynamic (S,s) economies. Econometrica 59, 1659-1686. Caplin, ., Spulber, ., 1987. Menu costs and the neutrality of money. Quarterly Journal of Economics 102, 703-725. Cukierman, A., 1980. The effect of uncertainty on investment under risk neutrality with endogenous information. Journal of Political Economy 88, 462-475. Deaton A., 1992. Understanding Consumption. Oxford University Press, New-York. Dorunbusch, R., 1987. Open economy macroeconomics: new directions. NBER Working Paper 2372. Eberly, ., 1994. Adjustment of consumers’ durables stocks: evidence from automobile purchases. Journal of Political Economy 102, 403-436. Eckstein, Z., Tsiddon, D., 2004. Macroeconomic Consequences of Terror: Theory and the Case of Israel. Journal of Monetary Economics 51(5), pp. 971-1002. Eldor R., Melnick, R., 2004. Financial markets and terrorism. Working Paper, Arison Business School, The Interdisciplinary Center, Herzliya. Enders, W., Sandler, T., 1991. Casualty between transnational terrorism and tourism: the case of Spain. Terrorism 14, 49-58. Enders, W., Sandler, T., 1996. Terrorism and foreign direct investment in Spain and Greece. Kyklos 49, 331-352. Enders, W., Sandler, T., Parise, ., 1992. An econometric analysis of the impact of terrorism on tourism. Kyklos 45, 531-554. Flavin, ., 1981. The adjustment of consumption to changing expectations about future income. Journal of Political Economy 89, 974-1009. 36
Fielding, D., 2003. Counting the cost of the Intifada: consumption, saving and political instability in Israel. Public Choice 116, 297-312. Fleischer, A., Buccola, S., 2002. War, terror, and the tourism market in Israel. Applied Economics 34, 1335-1343. Hall, ., 1978. Stochastic implications of the life cycle-permanent income hypothesis: theory and evidence. Journal of Political Economy 96, 971-987. Hall, ., 1989. Consumption. In: Barro . (Ed.), Modern Business Cycle Theory and Evidence, Harvard University Press, Cambridge, MA, 153-177. Hassler, J., 2001. Uncertainty and the timing of automobile purchases. Scandinavian Journal of Economics 103, 351-366. Ingersoll, ., Ross, ., 1988. Waiting to invest: investment and uncertainty. Unpublished paper, Department of Economics, Yale University. Lam, ., 1989. Irreversibility and consumer durables expenditures. Journal of Monetary Economics 23, 135-150. Leahy, ., Zeira, J., 2004. The timing of purchase and aggregate fluctuations. NBER Working Paper No. W7672. Mankiw, ., 1982. Hall’s consumption hypothesis and durable goods. Journal of Monetary Economics 10, 417-425. Newey, ., West, ., 1987. Simple positive semi-definite, heteroscedasticity and autocorrelation consistent covariance matrix. Econometrica, 703-708. Pindyck, ., 1991. Irreversibility, uncertainty, and investment. Journal of Economic Literature 29, 1110-1148. Romer, ., 1990. The great crash and the onset of the great depression. Quarterly Journal of Economics 105, 597-624. 37
