Policy Research Working Paper 10991 The Only Way Is Up? Economic Mobility in Malaysia in the 21st Century Gerton Rongen Peter Lanjouw Poverty Global Department December 2024 Policy Research Working Paper 10991 Abstract This study documents short-term economic mobility in progress is not uniform: around 15 percent of the popula- Malaysia over the first two decades of the twenty-first cen- tion in rural East Malaysia lives in chronic poverty. Second, tury, at the population level and for various subgroups. The the study finds considerable increases in sustained economic findings show broad and steady improvements in well-being, security—the extent of improvement, however, depends on as evidenced by large decreases in chronic poverty and sig- the approach and income thresholds that are used to define nificant increases in persistent economic security. The study security. Moreover, ethnic and regional differences in secure employs a synthetic panel approach based on nationally status are sizable at higher income class thresholds. The representative micro-level data for 2004–22, with a refine- largest differences are of a regional dimension: an individ- ment that allows presenting bootstrap point estimates and ual in urban Peninsular Malaysia is more than three times standard deviations. In addition, the study investigates sev- more likely to live in economic security than someone in eral poverty and vulnerability scenarios, as well as relative rural East Malaysia. Altogether, the study observes upward mobility. First, the results indicate that chronic poverty has movement across the board but little evidence of dramatic decreased to 2–3 percent of the population. Nevertheless, changes in the relative positions of societal groups. This paper is a product of the Poverty Global Department. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at a.g.m.j.rongen@vu.nl. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team The Only Way Is Up? Economic Mobility in Malaysia in the 21st Century Gerton Rongen*a and Peter Lanjouwa JEL codes: C53, D31, D63, I32, O53 Keywords: Poverty dynamics, income mobility, ethnic inequality, regional inequality, Malaysia, synthetic panels This paper was commissioned by the World Bank’s Poverty and Equity Global Practice, East Asia and the Pacific Unit (EEAPV) for the report A Fresh Take on Reducing Inequality and Enhancing Mobility in Malaysia under the project Malaysia Equity and Inclusion (FY23-25). The report was led by Laura Rodriguez, Ririn Purnamasari and Matthew Wai-Poi. Acknowledgments: the authors would like to thank Chris Elbers for his invaluable suggestions, as well as Laura Rodriguez, Ririn Purnamasari and Matthew Wai-Poi for their comments and reflections. * Corresponding author: a.g.m.j.rongen@vu.nl a Vrije Universiteit Amsterdam; Amsterdam Institute for Global Health and Development 1. Introduction This study analyzes poverty dynamics and economic mobility in Malaysia over the period 2004- 2022. We make use of a synthetic panel based on nationally representative cross-sectional microdata to investigate the movements of Malaysian households across the income distribution. We take three perspectives: first, we study mobility at the lower end of the distribution by focusing on poverty dynamics. Second, we take a broader view of economic mobility by dividing the distribution into three groups and analyzing upward and downward mobility along those dimensions. Finally, we assess relative mobility by considering transitions across quintiles over the entire income spectrum. This study builds on the findings of Rongen et al. (2024) in four ways. First, it extends the analysis by including survey data on 2019 and 2022. Second, it takes an expanded view by explicitly analyzing various scenarios for setting the poverty line and for incorporating the concepts of vulnerability and economic security. Third, it gives the first estimates of relative mobility based on these microdata. Lastly, we introduce a methodological refinement such that we are able to present bootstrap point estimates for all transitions, whereas the previous study estimated upper and lower bounds on mobility only. We conclude that the picture of economic mobility in Malaysia over the period 2004-2022 is generally positive, with large reductions in chronic poverty and a growing group that we classify as economically secure. However, these gains are not uniform and depend on how we define income groups, and on which population subgroup we consider. Chances of remaining poor and of falling into poverty are highest in rural East Malaysia. Moreover, attaining economic security through sustained upward mobility is particularly difficult in rural East Malaysia and for East Malaysia Bumiputeras, two groups that overlap to an important extent. Downward mobility seems to have increased for these groups. Lastly, relative mobility results show considerable persistence in the position of sizeable groups at either the bottom or the top of the distribution. Over our study period, we do not observe episodes of greater relative mobility – the ordering of Malaysian society as expressed through its income distribution did not exhibit major changes. The paper proceeds by discussing methodology and data in the next section. Thereafter, we provide results in section 3, which consists of three subsections. We draw some conclusions in the final section. 2. Methodology and data 1 2.1. Bound estimates for poverty dynamics This section summarizes the approach presented in Dang et al. (2014). Our aim is to estimate joint and conditional probabilities of poverty dynamics. For example, we wish to know the likelihood that an individual is poor in survey round 1 and non-poor in round 2, a joint probability which we can write as follows, denoting per capita income by and the poverty line by : (1 ≤ 1 2 > 2 ). Alongside evidence on joint probabilities, we are also interested in conditional probabilities. For instance, we might wish to estimate the likelihood of moving out of poverty, conditional on being poor initially. Such a conditional probability can be written as (1 ≤ 1 2 > 2 ) (2 > 2 | 1 ≤ 1 ) = (1 ≤ 1 ) and can be computed straightforwardly from the joint probabilities, as the right-hand side of the equation shows. Panel data are not available in Malaysia. In the absence of such data, we do not observe incomes in round 1 and round 2 for the same unit and would normally not be able to estimate these probabilities. However, the method we describe below draws on cross-sectional data and yields both upper and lower bounds for each of the four possible outcomes of the dynamic process, in both conditional and unconditional terms. We model the logarithm of per capita income in each round as a linear function of time- invariant household characteristics and an individual error term. Schematically: ′ 1 = 1 1 + 1 and ′ 2 = 2 2 + 2 . The basic idea of the method is to use time-invariant household characteristics to predict household per capita income for the year in which the actual observation is lacking. Such characteristics will be 1 This section is largely based on the description of methodology in Rongen et al. (2022). 1 unchanged across the two rounds. Mostly, they will be characteristics of the household head, such as the year of birth, educational attainment, ethnicity or religion and birth district. In addition, if the survey includes retrospective questions, these could also be included, e.g., if the household owned a car or television at the time of the previous survey. As we have data for each survey round, we estimate the coefficient vector on these time-invariant explanatory variables. In practical terms, we predict the round 1 income for households in the round 2 survey by multiplying the coefficient vector resulting from the round 1 regression ( 1 ) with the characteristics of households in round 2 (2 ). This prediction looks as follows for the observations of round 2, with the addition of an error term estimate: � �1 = ′ 1 2 + ̂1 . We turn first to the assumptions underlying this approach and will then discuss how estimates for the error term are obtained. The method for estimating bounds on poverty transition quantities is predicated on two important assumptions. First, we need to assume that the underlying population from which the data are sampled is the same in both survey rounds. This implies that the method cannot be applied in cases where there are large shocks to the population, for example, if there were massive migration into or out of the country. More precisely, we want to ensure that the income model coefficients that we estimate in round 1 are good predictors for what the round 1 income of the households surveyed in round 2 would have been. Hence, to ensure stability of the reference population, we only include households whose heads are between 25 and 60 years in survey round 1. 2 Second, we need to make assumptions about the relationship between the error terms 1 and 2 . Specifically, we assume that these are, on average, positively correlated over time. In the presence of household-specific effects and persistence of shocks, this does not seem to be an unreasonable assumption. In settings where panel data are available, the correlation coefficient of log income can be estimated directly, and indeed studies report values between 0.53 and 0.89 across a variety of countries (Dang et al. (2014); Colgan (2023, 2024). We do not make further assumptions about the shape of the distribution of the error terms for these bound estimates; in that sense, the resulting bounds on poverty transitions are non-parametric. The upper- and lower-bound estimates for mobility derive from two extreme cases of the assumed relationship between the error terms. It is important to note that upper bounds of mobility 2 This age range is chosen so that formation of new households and dissolution of existing households should be at a minimum. Moreover, it ensures that in round 1 heads will generally have obtained their maximum level of education. 2 correspond to the lower bound of immobility, i.e., remaining poor or remaining non-poor, and vice versa. Throughout the paper, upper-bound estimate and lower-bound estimate refer to upper and lower bounds of mobility. At one extreme, we consider the case of = 0, which corresponds to the upper-bound mobility estimate. In this scenario, the errors are uncorrelated and mobility, given household characteristics, will be at a maximum, both out of poverty and into it. Because we assume that errors in fact are positively correlated, on average the upper bound will be strictly larger than actual mobility. Predicted upper-bound round 1 income for household surveyed in round 2 will be: �2 � ′ � 1 = 1 2 + ̂1 , � 1 is a residual randomly drawn from the actual distribution of residuals resulting from where ̂ estimating the income model on the round 1 data. 3 Based on the predictions thus obtained, we can estimate the probability of each poverty transition for each observation. Aggregating these quantities at the population level, we obtain the country-level poverty dynamics. 4 Due to the random nature of this process, these steps are repeated R times, with R = 100 in this application, and averaged to obtain a robust estimate of the upper bound of mobility. At the other extreme, our lower-bound estimates assume a perfect positive correlation between the errors: = 1. In this case, average mobility will be at its minimum. For the estimation, this means we simply take the estimated round 2 residual, scale it by , and add it to the fitted round 1 income: 5 �2 � ′ 1 = 1 2 + ̂2 . Analogous to the upper-bound scenario, but without the repetition, we obtain the lower-bound population-level poverty dynamics estimates. Conditional probabilities are subsequently computed from the population- or subgroup-level joint probability estimates. 3 In contrast to Rongen et al. (2024), we do apply population weights in drawing this random sample of residuals, in response to a discrepancy we found between the previously estimated bounds and proposed bootstrap point estimates that we discuss further down. The resulting changes to bound estimates are minor, and mobility trends remain unaffected. 4 In our analysis the regressions are at the household level taking log per capita household income as the dependent variable. Transition probabilities for each household are population-weighted when we aggregate, so that the country-level figures represent population proportions. 5 We use a scaling factor = �2 to adjust for the difference in error variance between the two rounds ( �1 ⁄ � is the standard error of the residuals). 3 2.2. Bootstrap point estimates We propose a Bayesian-inspired innovation to the synthetic panel method that allows us to present point estimates of transition quantities. 6 Our point of departure is as follows: we do not know the actual degree of household income correlation in Malaysia, but several panel studies have documented such correlations over time for a range of countries. We assume that the true values for Malaysia will be somewhere in this range of empirical correlations. Our proposition, then, is that a bootstrap sample of these empirical income correlations, coupled with the assumption that residuals follow a bivariate normal distribution, yields a meaningful point estimate for transition quantities, and plausible standard deviations. Obviously, we do not know what the actual Malaysian correlations are in each interval, so that we will focus on the bandwidth around the point estimate, rather than the actual estimate itself. Our point estimates are based on the method put forward by Dang et al. (2014), the same paper that introduced bound estimates. Precision comes at the cost of introducing additional assumptions, however. Notably, we assume that the errors of our income model, 1 and 2 , follow a bivariate normal distribution characterized by the correlation coefficient . We derive hypothesized values for the correlation of the error terms () by drawing on income correlation coefficients (12 ) from actual panel data in other countries. Dang and Lanjouw (2023) provide the following proposition for doing so: ′ 1 2 �(1 ) (2 ) − 1 ( )2 = �(1 ) (2 ) The equation combines measures of the variation in our data with a hypothetical income correlation to arrive at an estimate of . Our sample of the income correlations has been obtained from Colgan (2023, 2024) and Dang and Lanjouw (2023), who provide actual panel correlations for fourteen countries in the EU-SILC database and the United States. We standardize these empirical correlations to a two-year reference interval, since the correlations reported covered intervals of different duration. Subsequently, we draw a bootstrap 6 Dang & Lanjouw (2023) also present a method to obtain point estimates, which depends on cohort-level income correlations to infer household-level correlation. However, the robustness of this method is not generally accepted. Instead, we make use of documented empirical measurements of household-level correlation in other countries. 4 sample of size 100 from these standardized correlations. 7 For each draw, we estimate transition probabilities for all intervals of adjacent survey rounds in our data. By interval, we first compute the resulting error correlation, and then employ the bivariate standard normal cumulative distribution function 2 (. ) to obtain estimates for all poverty transition probabilities, e.g. as follows for the poor to poor transition8: � 1 − � 1 ʹ2 2 − 2 ʹ2 � (1 < 1  2 < 2 ) = 2 � , �� , � 1 � 2 For each draw, household level results are aggregated up to population and subgroup level. Then, averaging over all 100 draws yields the bootstrap mean probabilities, our point estimates, and allows us to compute standard deviations over the 100 draws. 2.3. Vulnerability We analyze vulnerability and the achievement of economic security in two separate ways. One approach is based on the definition of a vulnerability line as introduced by Dang and Lanjouw (2017). Below, we summarize this approach and explain how it ties in with the synthetic panel method. Our goal is to estimate 3x3 transition matrices for the income categories poor, vulnerable and economically secure.9 This approach focuses on the lower part of the income distribution, which is reflected in our choice of income categories. We note that once an individual reaches the secure category, there is no further scope for mobility in this set-up. Further down, we outline two other scenarios that are able to shed light on mobility higher up the income distribution. We start by defining the category ‘vulnerable’ as those non-poor in period 1 who face a certain probability or higher of falling into poverty in period 2. This conditional probability can be denoted as: (2 ≤ 2 1 < 1 ≤ 1 ) (2 ≤ 2 | 1 < 1 ≤ 1 ) = , ( 1 < 1 ≤ 1 ) 7 We also used maximum likelihood estimation to fit beta and normal distributions to the empirical data, from which we then would have been able to draw a sample of coefficients. However, the fit of these theoretical distributions was not completely satisfactory, such that we preferred to stick with the empirical distribution. 8 We refer to Equations 20-23 in Dang et al. (2014) for all transition formulae. 9 There are no generally agreed upon names for these categories. In some applications, the non-poor who are not considered vulnerable have been labelled ‘middle class’. We opt for the more neutral ‘economically secure’. One reason is that some of these households will be quite rich, so that middle class would be a misnomer for them. 5 where 1 stands for the vulnerability line. The latter is derived empirically at population level after setting the vulnerability index , the highest probability of falling into poverty that we are willing to accept before we no longer consider a household to be economically secure. For example, we might set the index at a value of 20 percent. If a household faced a probability of falling into poverty greater than 20 percent, we would consider this household to be vulnerable. Formally, the vulnerability line is then derived from this index as follows 10: max { 1 | (2 ≤ 2 | 1 < 1 ≤ 1 ) ≥ } This approach contrasts with other approaches in the sense that the vulnerability line is not set arbitrarily, for example at 1.5 times the poverty line, but derived empirically from the vulnerability index. In practice, this requires a recursive algorithm that goes over the synthetic panel data to arrive at the vulnerability line that yields the desired vulnerability index. We acknowledge that setting the vulnerability index is still an arbitrary decision, but it is based on a transparent weighing of risks that evokes the insecurity that is inherent to vulnerability. We mostly present results for poverty and vulnerability dynamics in figures that do not indicate the level of precision of the estimates. The main reason for this is that the results already consist of bounds that incorporate uncertainty about the errors in our income model. The bounds are highly likely to contain the true value of the probability estimates, as was confirmed in other settings by Dang et al. (2014) and Hérault and Jenkins (2019). In addition, sample sizes for each round are large, so that the standard errors of our estimators are generally small. The second approach is based on a method for analyzing vulnerability introduced by Chaudhuri (Chaudhuri et al. (2002); Chaudhuri (2003)). 11 Here, we again conceive of vulnerability as a certain probability of future poverty, with the important feature that a household that is currently poor may not necessarily be vulnerable: its characteristics could make the likelihood of poverty in the next period low, but it could have an income below the poverty line this period due to some shock. A household’s probability of future poverty, i.e. its Chaudhuri vulnerability, is estimated as follows: ̂ ln − ℎ � (ln ℎ < ln | ℎ ) = � �ℎ = Pr � � �ℎ where it is assumed that the error term of the regression follows a normal distribution. The probability that household income ℎ will be below the poverty line , given household characteristics ℎ , is then 10 Recall that we are assuming a positive correlation between the errors in our income model, which ensures that (2 ≤ 2 | 1 < 1 ≤ 1 ) is decreasing in 1 . 11 One reason for including this approach is that the World Bank uses it across the East Asia Pacific region to define vulnerability lines based on a consistent methodology, see Krah et al. (forthcoming). 6 derived from the standard normal cumulative density function (. ), with estimated expectation and ̂ and ℎ variance of ℎ � respectively. We apply this method primarily to derive aggregate income thresholds that separate the following three groups: the poor and vulnerable, the aspiring middle class, and the middle and upper class. 12 Note that the grouping in this scenario differs considerably from the classification in the previous scenario. Let us call the threshold between the first two groups in this scenario the Chaudhuri vulnerability line, and the line dividing the last two groups the middle class line. The Chaudhuri vulnerability line is the average predicted per capita income of the group of households that have a certain probability of being poor in the future, 10 percent in our application. Building on this, the middle class line is the average predicted income of households with a certain probability of being vulnerable, again 10 percent in our application. Households are then classified into the three groups according to these two lines. 2.4. Relative mobility So far we have detailed how we analyze movements past certain fixed monetary goal posts. A limitation of such absolute mobility analysis is that what happens once households move comfortably beyond such thresholds is out of our analytical scope. Analyzing relative mobility across the five quintiles of the income distribution enables us to take a broader view and to consider movements along the entire income spectrum. Quintile thresholds are determined based on the cross-sectional income distribution in a given year, so they differ from year to year. The synthetic panel is then employed to estimate the probability that a household has made a particular transition from one quintile to another over adjacent survey rounds. We obtain a 5 x 5 transition matrix once estimates are aggregated. For the point estimates of relative mobility, we apply the method as described above, substituting the quintile thresholds for the poverty lines in the bivariate standard normal cumulative distribution function above. For the bounds, we take a different approach from the one described in section 2.1. Because we want to estimate the probabilities of twenty-five transitions we need higher precision: the standard bounds approach would likely result in uninformative estimates. Hence, we take the smallest and largest correlation values in our sample of income correlations to estimate our relative mobility bounds. 12 We use this terminology in line with the umbrella study on inequality in Malaysia to which this paper contributed. 7 The lowest correlation coefficient in our sample is 0.53 for Greece, while the highest is 0.89 for Czechia. 13 These represent respectively the high and low mobility scenario, as the closer the coefficient is to one, the stickier incomes are. A further assumption then is that these imported coefficients are sufficiently extreme to form upper and lower bounds in the case of Malaysia. Given the variety of countries included, at various levels of economic development, we suggest that this assumption is not unreasonable. The relevant transition probabilities can then be estimated by drawing on the bivariate standard normal cumulative distribution function 2 (. ), as before. For example, the probability of being in the bottom quintile two surveys in a row is estimated as follows: 20 1 − � 20 � 1 ʹ2 2 − 2 ʹ2 � (1 < 1 20 20  2 < 2 ) = 2 � , �� , � 1 � 2 20 where represents the 20th percentile of the distribution in year and the other symbols are as defined in section 2.1. 2.5. Data This study analyzes eight rounds of cross-sectional data from the Household Income Survey (HIS) collected by the Department of Statistics Malaysia. The data are for the years 2004, 2007, 2009, 2012, 2014, 2016, 2019 and 2022; we consider all pairs of adjacent survey rounds. 14 One requirement for analyzing welfare changes over time is that the measure of welfare, income in our case, is measured in a consistent and comparable manner. The Malaysian data are of high quality and satisfy this requirement. In addition, the sample sizes of the surveys are large, ranging from 36,000 to over 80,000 household-level observations per survey year. The dependent variable in our analysis is the natural logarithm of monthly household pre-tax income per capita, which is evaluated against our poverty line to determine the poverty status of a household. We did not use post-tax income because few Malaysians are subject to personal income tax; it is thus unlikely to affect poverty or vulnerability status. Households report income over the twelve months preceding the interview. Notably, this means that households interviewed in 2022 also report over part of 2021. 13 In the case of Greece, this is the correlation of disposable household income between 2001-2004; for Czechia, it is the 2014-2015 correlation of disposable incomes. 14 Surveys may extend into the subsequent year. When we refer to ‘round 1’, this indicates the first round of any pair, not necessarily the 2004 survey (and similarly for ‘round 2’). 8 Income data have been adjusted for spatial price differences. We constructed a spatial price index that reflects prices of basic goods, services and housing at the state level, separately for rural and urban areas within the state (see Table A in the Appendix). The index for the years 2004-2016 was derived from the household-specific poverty lines that the Department of Statistics Malaysia calculates for each survey year, which depend on the location and demographic composition of the surveyed household. 15 For 2019 and 2022, we extrapolated the index based on the CPI of the corresponding area. As explanatory variables in our income model, the method requires we only use variables that are time-invariant. We use three such categories. First, we employ dummy variables for the birth cohort of the household head, using five-year bins. The first bin starts in 1940; the cohort 1965-1969 is the omitted dummy, since it has members in every survey year. Second, we use six dummy variables for the highest level of education enjoyed by the household head. Those with primary education or less form the base category. Third, we use dummy variables for the ethnicity of the household head. The four categories are Bumiputera, Chinese, Indian and Other; Bumiputera is set as the base category. Fourth, we use a dummy variable for the gender of the household head. In addition, we use two location dummies that, strictly speaking, are not time-invariant. These are one dummy for urban or rural location and another for being located in either Peninsular Malaysia or East Malaysia. This restricts the interpretation of results to households that did not migrate from one such classification to another. Table B in the Appendix provides the OLS regression results of our income models for each survey year, based on the subsample of households with heads in the age range of 25–60. 3. Results We discuss the results of our analysis of economic mobility in three parts. First, we focus on absolute mobility at the lower end of the income distribution by investigating poverty dynamics. Then, we take a broader perspective by studying upward and downward absolute mobility between three income groups, which takes into account considerations of vulnerability and economic security. These first two analyses are based on fixed income thresholds. The third part studies relative mobility by analyzing movements between income quintiles. This allows us to analyze what is happening in the richer parts of the income distribution too. 15 We are aware that the index is thus based on expenditure patterns at the very low end of the income distribution. However, this is the best we could do in the absence of spatially disaggregated consumer price data. 9 Figures in this section present both bound estimates and bootstrap point estimates, as well as a two standard deviations band (both plus and minus) around the point estimate. In discussing results, we will focus on this band, rather than the point or bound estimates. 3.1. Poverty dynamics In this section, we analyze poverty transitions over adjacent survey rounds in the period 2004 -2022. We have four transition outcomes: chronic poverty, poverty exit, poverty entry, and not being poor in two rounds in a row. Chronic poverty is defined as being poor in both rounds. We discuss the unconditional probabilities of being in each state, which are equal to estimated population shares (see also Table C in the Appendix). In addition, we discuss conditional probabilities, such as the probability of poverty exit, given that the household was poor in the first round. The conclusions in this section are based on three scenarios for setting the poverty line. These are: 1. The 40th percentile of the income distribution in 2004, at RM 527 per capita per month. This is the same line as used in Rongen et al. (2022). 16 We refer to this as the main scenario. 2. The per capita average of the household-specific poverty lines calculated by DOSM for the households in the 2019 sample, at RM 559 per capita per month. Lines are household-specific in the sense that they vary by location, household size and composition. 3. The same approach based on the 2022 household-specific poverty lines, at RM 618 per capita per month. In all scenarios, the value of the poverty line is held constant in real terms across the entire study period. This may be regarded as a drawback of our analysis, since aspirations and consumption patterns will have changed between 2004 and 2022: what is considered poverty in 2022 may not have been seen as such in 2004. However, we have opted for absolute poverty lines in order to better track developments over time. By extension, we should note that our estimates do not correspond to the official Malaysian poverty figures, which are based on poverty lines that change between surveys. Hence, our findings may appear surprising given that the DOSM Poverty Report (DOSM, 2023) shows that poverty has risen recently, from 5.6 percent in 2019 to 6.2 percent in 2022. However, this may be 16 Note that all lines are given in 2016 Malaysian ringgit, so they represent the purchasing power of the ringgit in that year. The poverty line used in scenario 1 was selected because discussions about poverty and mobility in Malaysia often refer to the living standards of the bottom 40 percent of the income distribution. When our previous study commenced, the official Malaysian poverty line was very low and poverty virtually non-existent by those standards. 10 ascribed to a large increase in the poverty line going from 2019 to 2022, particularly in the non-food element. If we apply either the 2019 or the 2022 poverty line and fix its real value over time, we observe a decrease in cross-sectional poverty from 2019 to 2022. Figure 1 shows our estimated population shares of the four poverty transition outcomes over time under the main scenario. Headline transition results are qualitatively the same, irrespective of which of the three poverty lines is used. 17 This means that all three scenarios show a consistent image of declining chronic poverty. In the main scenario, we observe a decrease from 22-28 percent of the population in chronic poverty over the 2004-2007 interval to a 2-3 percent share over 2019-2022. Moreover, the share of the population that is consistently non-poor increased to around 88 percent of the population. Over time, the share of the population that escapes poverty is decreasing, which is not surprising given that a much smaller share of the population lives in poverty. Similarly, the share of the population that falls into poverty has also decreased much, to between 3-4 percent of the population. Together, these result in much less chronic poverty. Further below, we investigate to what extent these results hold across different groupings of Malaysians. It should be noted that social assistance policies instituted in response to the Covid-19 pandemic will have influenced these results. For example, we could have hypothesized that the pandemic would have led to higher poverty entry figures. However, we do not observe such an outcome over the 2019-2022 interval. Obviously, we cannot draw any conclusions about the years in between, when the Covid-19 pandemic hit hardest. Still, it seems that Covid-19 did not lead to large- scale permanent poverty and it is likely that government relief programs have contributed to this. Total budgets of relief packages for households and businesses equaled 23 percent of GDP in 2020 and 15 percent in 2021 (World Bank, 2021). For example, a World Bank survey indicated that about 75 percent of households with monthly income below RM 4,000 had received cash transfers between June and November 2021 (World Bank, 2021). Even so, this leaves an important share of the eligible population without assistance. 17 For comparison, the same graph based on the highest poverty line (RM 618 under scenario 3) is given in the Appendix. 11 Figure 1 Poverty Dynamics: Population shares 2004 - 2022 Next, we turn to estimated conditional probabilities under our main scenario, as displayed in Figure 2. Note that we can also interpret these in a second way, e.g. the conditional probability of remaining poor equals the share of poverty that is chronic. The enhanced precision of the bootstrap estimates allows us to draw firmer conclusion about e.g. chances of escaping poverty compared to Rongen et al. (2022). We observe that, on aggregate, the non-poor face only a small chance of falling into poverty, between 3-5 percent, over the 2019-2022 interval. This has fallen from a probability between 13-21 percent at the start of the study period. This headline number masks considerable variation across subgroups, as will be discussed further down. For a poor individual in 2004, chances of moving out of poverty (29-41 percent) were considerably lower than chances of remaining poor (59-71 percent). However, this situation was reversed over 2019-2022, such that for poor individuals, chances of moving out of poverty (60-76 percent) are now clearly higher than those of remaining poor (24-40 percent). Results are qualitatively the same in the other scenarios. This implies that for most of the poor, poverty is now a transient experience. Still, a considerable share of the poor, let’s say one in three, will not be able to escape poverty. We need to recall that government transfers in 2021 and 2022 were significantly higher than in previous years. This will have influenced poverty outcomes. If policy eventually returns to usual in 12 the post-Covid period, it is not certain that this trend of growing chances of exiting poverty will continue. Figure 2 Poverty Dynamics: Conditional probabilities 2004-2022 Next, we consider subgroup poverty dynamics. The upper-left panel of Figure 3 shows that the decreasing trend in chronic poverty holds for all three main ethnic groups. However, clear level differences exist: over the 2004-2007 interval, we estimate rates to be between 30-35 percent for Bumiputeras, between 6-10 percent for Chinese Malaysian, and between 16-21 percent for Indian Malaysians. By 2019-2022, these rates had decreased to 3-5 percent, around 0.5 percent and 1-2 percent, respectively. Absolute differences in chronic poverty rates have become much smaller. Conditional poverty transition probabilities also converge over time in absolute terms, as can be seen in Figure 4. Chances of escaping poverty have increased for all main ethnic groups, as has the likelihood of staying out of poverty. Over 2019-2022, the former probability is around 58-74 percent for Bumiputeras, 76-91 percent for Chinese Malaysians and 70-86 percent for Indian Malaysians; the latter stands at around 95 percent, 99 percent and 97 percent, respectively. It seems that the 2009- 2012 period was a particularly effective time for escaping poverty. 13 Figure 3 Poverty transitions by ethnic group. The vertical axis indicates the share within the group. Figure 4 Conditional poverty transition probabilities by ethnic group 14 We continue our subgroup analysis by turning to chronic poverty in geographically based groups. As Figure 5 demonstrates, we observe notable differences between regions, which are most pronounced when we compare rates over the most recent interval in urban Peninsular Malaysia (about 1 percent; third panel from left) to those in rural East Malaysia (13-17 percent; rightmost panel). 18 Rates in rural Peninsular Malaysia are comparable to those in urban East Malaysia, at 3-6 percent. At the same time, it should be noted that chronic poverty has declined dramatically in rural East Malaysia since the start of our study period. Figure 5 Chronic poverty by geography. The vertical axis indicates the share within the group. We can also combine these geographical and ethnic dimensions, as in Figure 6. We observe that the contrast in chronic poverty rates between Peninsular and East Malaysia (panels on the left side of Figure 5) is repeated when we restrict the analysis to Bumiputeras living in those parts (first and third panels Figure 6). The difference is much less pronounced for non-Bumiputeras. Within East Malaysia, we observe pronounced differences in chronic poverty between ethnic groups, with the absolute gap narrowing over time. Still, around 8-12 percent of East Malaysia Bumiputera live in chronic poverty. 18 We do not show separate figures for urban and rural Malaysia – trends in these areas are comparable to those for Peninsular and East Malaysia respectively. 15 Figure 6 Chronic poverty by ethnicity and geography. The vertical axis indicates the share within the group. Another dimension of poverty to consider is the share of a subgroup that enters poverty in a given interval, as displayed in Figure 7. As documented above, this share has fallen over time for the general population and across the three main ethnic groups. However, across geographically defined groups, we do not observe such a uniform pattern. While poverty entry has clearly fallen over time for Peninsula-based groups, this is not the case for East Malaysia. Notably, there does not seem to be much change in this respect for rural East Malaysia and East Malaysia Bumiputera over our study period of almost two decades; for example, poverty entry in rural East Malaysia continues to hover between 6-14 percent. This is especially worrisome since our poverty line is an absolute one that is adjusted for price inflation only, not for greater overall affluence in Malaysia. So, while chronic poverty has fallen dramatically also in East Malaysia, its population continues to experience shocks that make falling into poverty a common phenomenon. 16 Figure 7 Poverty entry shares across geographic groups Differences in poverty dynamics between subgroups can also be inferred by analyzing conditional probabilities. For example, given that a person is in poverty, what are their chances of staying poor? In our setup, if a subgroup faces an elevated probability of remaining poor, that could well be indicative of the depth of poverty in that group, i.e. that the poor belonging to this group have incomes further below the poverty line than the average poor person. In addition, an elevated conditional probability indicates that, for this subgroup, poverty is more often chronic (rather than transient) than it is across the full population. Figure 8 gives the likelihoods of remaining poor and of falling into poverty over the 2019-2022 interval for a selection of subgroups. 19 In the left panel, we observe that those living in rural East Malaysia were most likely to remain poor if they were poor in 2019, at a probability of 40-55 percent, compared to a 24-40 percent mean probability. Rural areas in general and East Malaysia as a whole also face elevated probabilities, as do Bumiputera living in East Malaysia. For these groups, poverty is more often a chronic phenomenon than it is for Malaysians in other groups. For example, Chinese 19 We only discuss these two because the conditional probabilities are two sides of the same coin: the probability of exiting poverty equals one minus the probability of staying in poverty, while the probability of staying out of poverty equals one minus the probability of falling into poverty. 17 Malaysians were least likely to remain poor. Next, considering the likelihood of falling into poverty in the right panel, we learn that most subgroups face probabilities that are close in absolute terms to the mean probability of 3-5 percent. However, the same groups as before now face elevated probabilities of falling into poverty, indicating that the incomes of the non-poor in these groups are not as far above the poverty line as they are in other groups. Notably, non-poor people in rural East Malaysia face a 12- 19 percent probability of falling into poverty, which is about four times as large as the average. Figure 8 Likelihood of poverty transitions (2029-2002) across subgroups In sum, we observe large decreases in chronic poverty over 2004 to 2022. Moreover, those living in poverty have a better chance of escaping poverty now than in the beginning of that period. Our subgroup analysis has shown that these developments are broad-based, although important differences remain. It is not at all obvious that economic growth will have such a broad reach – one can imagine scenarios in which, for example, rural or more remote areas do not benefit much. In Malaysia, however, absolute gaps in poverty rates between population groups have become smaller in most cases. Of the various subgroups that we studied, chronic poverty rates are most elevated among the population in rural East Malaysia. In addition, poverty entry remains frequent there. By extension, 18 this means the population in this area faces the highest chances of remaining poor when poor and of falling into poverty when not poor. 3.2. Absolute three-way economic mobility In this section, we go beyond a two-way poor/non-poor classification and split the income distribution into three groups, incorporating vulnerability as an analytical concept. This enables us to study absolute mobility for a larger group of Malaysians. We focus on those who experience either upward mobility (moving up one or two groups between surveys), downward mobility (moving down one or two groups), persistent security (remaining in the richest group), or persistent vulnerability (stuck in the middle group). We employ two different scenarios for this three-way classification. 20 Researchers have used various ways of defining the concepts of vulnerability and middle class, we apply two such approaches in this study. It is important to note that these scenarios depart from different analytical positions, such that results have somewhat different interpretations. In order to sketch what part of the income distribution we are considering, Figure 17 in the Appendix shows how the two scenarios map onto the distributions in 2004 and 2022. The scenarios are as follows: 1. A division into poor, vulnerable and economically secure groups, as described in section 2.23. Here we conceive of vulnerability as an elevated probability of falling into poverty. This employs the poverty line of poverty scenario 1 at RM 527 and a vulnerability line of RM 1,120. The latter is estimated by applying the method proposed by Dang & Lanjouw (2017) to the 2004-2007 transition interval and selecting a vulnerability index of 0.3333. It means that those with an income of between RM 527 and RM 1,120 in 2004 faced a one in three chance of having an income below RM 572 in 2007. 2. A classification where we derive the income thresholds by applying the Chaudhuri method (see section 2.3). Based on a 10 percent probability of future poverty, this yields a Chaudhuri vulnerability line of RM 920 and a ‘middle class line’ of RM 1,590. 21 The low-income group consists of the poor and vulnerable, the middle group is labelled ‘aspiring middle class’, and those with incomes above RM 1,590 cover the middle and upper classes in economic terms. By coincidence, RM 1,590 is also approximately the mode of the income distribution in 2022. 20 Note that all income levels are again in 2016 Malaysian ringgit, in per capita terms per month. 21 The poverty line here is the average of the official 2019 household-specific poverty lines (our scenario 2 poverty line). We also investigated results based on a 20 percent maximum probability of poverty (at thresholds of RM 742 and RM 1,063, respectively). This gave qualitatively similar results, so we do not report on the outcomes. 19 We present the scenario 1 mobility trends at the population level in Figure 9. Over the time period studied, we observe a large increase in the share of the population that is persistently secure, from 17 – 21 percent over 2004-2007 to between 50 - 55 percent over the 2019-2022 interval. In this scenario, we should interpret persistent security as having a small probability of falling into poverty. 22 At the same time, the share of the population that experiences downward mobility decreases somewhat, from between 15 – 21 percent to between 11 – 16 percent. Upward mobility was mostly at a level of around 20 percent of the population, with a small peak over the 2009-2014 period. The share of the population that is vulnerable two surveys in a row is sizeable, at between 14 – 19 percent until 2016. Most recently, the share seems to have decreased, to 11 – 14 percent. Given continued upward mobility and a small decrease in downward mobility, this seems to imply that more households are managing to attain incomes above our RM 1,120 per capita vulnerability line. Figure 9 Economic mobility - scenario 1 However, this image is different when we consider mobility under scenario 2 (Figure 10), in which the income boundaries lie at much higher levels. Hence, we are now focusing on movements in the middle of the income distribution. We still observe a significant increase in the share of the population that we classify as persistently secure (with incomes above RM 1,590 per capita in this scenario, labelled middle or upper class), from 8 – 11 percent to 31 – 36 percent of Malaysians. However, downward mobility under this scenario does not fall, and might even be higher at the end 22 Those who were classified as secure in 2004 faced an average conditional probability of between 2 – 10 percent to be in poverty in 2007; this had dropped to between 0 - 2 percent over the 2019-2022 interval. 20 of the study period: it stood at 13 – 17 percent over 2004-2007, but at 15 – 21 percent over 2019- 2022. Over this last interval, our estimates show that about 10 percent of the population experiences income drops from above to below our RM 1,590 middle class threshold. 23 In addition, a substantial group is stuck in the ‘aspiring middle class’ group, at 9 – 12 percent. On the positive side, upward mobility has also increased under this scenario, to about 25 percent of Malaysians. Taken together, there is more economic mobility in this second scenario, both upward and downward. This is not surprising since the per capita incomes of many Malaysian households are close to the income thresholds in this scenario; recall that our middle class line of RM 1,590 is close to the mode of the income distribution in 2022 (see Figure 17 in the Appendix). Elevated levels of downward mobility imply that for those in the middle of the income distribution, gains in one period are not necessarily permanent. Nevertheless, in both scenarios substantial groups reach persistent economic security. In addition, the two figures demonstrate that the bootstrap point estimate bands allow us to draw sharper conclusions, especially when it comes to upward and downward mobility, compared to the previous bound estimates. Figure 10 Economic mobility - scenario 2 Next, we turn to investigating mobility by subgroup. When we compare ethnic groups, we observe that there are clear level differences in persistent economic security, and that these have 23 Note that there is more scope for downward mobility now than at the start of the study period, simply because of the fact that many more people have obtained incomes above our income thresholds (see also Figure 17 in the Appendix). In our setup, there is no possibility of downward mobility if the household is in the lowest income group. 21 somewhat widened over time in absolute terms (Figure 11, bottom row). Subgroup shares increase from 10 - 14 percent to 43 - 47 percent for Bumiputeras, while the respective estimates are an increase from 37 – 39 percent to 76 – 78 percent for Chinese Malaysian and from 17 – 22 percent to 54 – 59 percent for Indian Malaysians. Nevertheless, a closer look reveals that the difference has shrunk in relative terms. For example, if an Indian Malaysian was about 60 percent more likely than a Bumiputera to be secure over the 2004-2007 interval, that had decreased to 25 percent more likely over the most recent period. Trends in upward and downward mobility also differ to some extent. After 2012, we see falling upward mobility for Chinese and Indian Malaysians, likely due to a ceiling effect (once the secure group is attained, one cannot go up further in our setup). However, upward mobility remains steady for Bumiputeras at around 20 percent. Downward mobility over the 2019-2022 period was lowest among Chinese Malaysians (6 – 9 percent), while more frequent among Bumiputeras (12 -18 percent) and Indian Malaysians (10 – 15 percent). Persistent vulnerability has decreased clearly for Chinese Malaysians, and slightly for Indian Malaysians, but remains at a rather steady level for Bumiputeras. Absolute mobility estimates under scenario 2, in which we apply higher group income thresholds, do not differ dramatically from results under scenario 1 (see Figure 18 in the Appendix). Clearly, the group shares of those who reach persistent security are much lower in this scenario. In tandem, we observe higher levels of upward and downward mobility overall. 22 Figure 11 Mobility by ethnic group - scenario 1 Turning to geographic regions, we focus on comparing urban Peninsular Malaysia to rural East Malaysia, as these exhibit the largest contrast. Figure 12 displays mobility trends under scenario 1; we 23 observe that persistent security is much more common in urban areas on the Peninsula. The share of the population here that is persistently secure has increased from 25 – 30 percent over the 2004-2007 interval to 60 – 64 percent over 2019-2022, against an estimated increase from 2 - 3 percent to 12 – 16 percent in rural East Malaysia. Over our study period, the absolute difference has widened to almost 50 percentage points; an individual in urban Peninsular Malaysia is more than three times more likely to live in economic security than someone in rural East Malaysia. In addition, both upward and downward mobility have increased in rural East Malaysia. One potential explanation is that a portion of households will have moved up one or two income groups in roughly the first half of our study period, only to move down again later. This observation would also be consistent with the picture of a more dynamic local economy, which, however, does not bring durable economic security. In addition, we observe high levels of persistent vulnerability, at 21 – 25 percent of the population. Qualitatively similar results under scenario 2 are presented in Figure 19 in the Appendix. Figure 12 Mobility by region - scenario 1 When we combine the geographic and ethnic dimensions and compare Bumiputeras in Peninsular Malaysia with those in East Malaysia, we observe a pattern similar to the geographic 24 differences discussed above, with the qualifier that the regional gap between those of Bumiputera ethnicity is less pronounced than the purely geographically-based gap. For example, under scenario 1, 48 - 53 percent of Bumiputeras on the Peninsula were persistently secure over the 2019-2022 interval, while this estimate amounted to 21 - 25 percent of East Malaysia Bumiputeras. In conclusion, we note that the picture of absolute economic mobility that we obtain depends on how we define the income groups. In both scenarios presented, we observe a steady pace towards more and more Malaysian attaining what we call persistent security (in scenario 1) or middle/upper class status (in scenario 2). However, these levels depend on the scenario and are higher if we set the group thresholds at lower levels of income. Notably, some groups are far behind the average when it comes to attaining persistent security, most prominently those living in rural East Malaysia. Another conclusion is that overall economic mobility, both upward and downward, is higher under scenario 2. Notably, under this scenario, downward mobility seems to be higher at the end of our study period for groups with lower initial incomes, specifically households in rural Malaysia, rural East Malaysia, and East Malaysia Bumiputeras. This also holds to some extent for the Bumiputera group as a whole. We would argue that for a nonnegligible part of the households concerned, such downward mobility in the later part of our study period is likely to have happened after prior upward mobility. Importantly, we should take into account that there is a so-called floor effect for groups with lower initial incomes: if a household is in the lowest income group, there is no scope for downward mobility in our setup. 3.3. Relative mobility In this section, we consider the entire income distribution by analyzing movements between income quintiles. This allows us to draw conclusions about the economic mobility of society as whole, albeit on a short-term timescale. Results can be represented in 5 x 5 transition matrices (see Table D in the Appendix). Figure 13 shows trends at the aggregate level for four categories of transitions. In general, we can observe that mobility patterns do not exhibit dramatic changes over time. By and large, the estimated shares of the population that are upward or downward mobile are rather stable. The left-most panel shows that approximately 24 - 32 percent of the population is upward mobile in any given interval, while the share of those moving down is generally between 27 – 34 percent (the second panel from the left). Of course, we should keep in mind that these are relative results: it could also be that an individual moves down because the incomes of others are growing faster. If incomes were random, we would expect to see 40 percent of the population moving up and another 40 percent moving down. 25 In addition, we observe that the shares of the population that remains in the poorest and richest quintiles are elevated compared to the population share of other quintiles. The third panel from the left shows that between 9 – 12 percent of the population remains in the bottom quintile, likely for prolonged periods of time. Some of these households will also be classified as chronically poor. On the other end of the distribution, those persistently in the top quintile amount to around 11 – 14 percent of the population, which indicates that the rich are rather successful in maintaining their advantage. Again, if incomes would be randomly drawn, only four percent of the population would be in each of these transition categories. However, studies have shown that individual advantages and disadvantages have a tendency to accumulate over the life cycle, such that this is not a surprising result (see e.g. Deaton & Paxson (1994)). Dang and Lanjouw (2023) report actual panel 5 x 5 transition results for Viet Nam over the 2006-2008 interval, which we may use for comparison. Our results for upward and downward mobility seem somewhat higher than these Vietnamese point estimates, which are 25.3 and 26.8 percent respectively. The shares of those remaining in the bottom and top quintiles in Viet Nam (at 12.7 percent and 12.9 percent respectively) are reasonably close to our estimates. Figure 13 Relative mobility between income quintiles These patterns at the aggregate level mask considerable variation at the subgroup level, notably at the top and bottom of the distribution. Note that the quintile income thresholds are still defined at the population level, they do not change across subgroups. Let us first consider levels and trends by ethnic group: we do not observe stark differences in overall mobility, although both upward 26 and downward mobility tend to be somewhat lower for Chinese Malaysians, which implies that this group is somewhat more likely to be immobile, i.e. to be in the same quintile across survey rounds. Notably, over the 2019-2022 interval, 40 - 50 percent of Chinese Malaysians remain in the same quintile, while this holds for just 33 - 43 percent of Bumiputeras. We can study this in more detail by zooming in on the proportion of each ethnic group that remains in the top quintile (the bottom panels in Figure 14). Between 22 - 29 percent of Chinese Malaysians is persistently in the top quintile, while this holds for around 10 - 14 percent of Indian Malaysians and for 7 - 10 percent of Bumiputeras. This situation is reversed when considering the proportion of each group in the bottom quintile, as shown in the top panels of the same figure. Figure 14 Persistence at the bottom and top by ethnic group Turning to subgroup mobility when we define groups based on their location, we again observe only small differences in overall patterns. One such difference is that overall mobility in rural East Malaysia seems to be lower than on the peninsula. As before, those remaining in the same quintile are concentrated either in the top or the bottom quintile, as Figure 15 shows. The panels on the left of that figure make clear that immobile households in rural East Malaysia are overwhelmingly in the bottom quintile: 36 - 42 percent of rural East Malaysians were persistently in that quintile over the 2019-2022 interval. On the other hand, only about 1 percent of this group is consistently in the top. In 27 contrast, those in urban Peninsular Malaysia are much less likely to remain in the bottom, and much more to stay at the top. This pattern of regional difference is repeated to a lesser extent when we compare Bumiputeras in East Malaysia to those in Peninsular Malaysia, on the right side of the figure. Figure 15 Persistence at the bottom and top by region and Bumiputera ethnicity 4. Conclusion This study has analyzed economic mobility in Malaysia over the period 2004 to 2022, based on nationally representative household survey data. We used a synthetic panel approach to enable us to draw conclusions about income mobility over time. First, we investigated poverty dynamics based on a simple poor/non-poor classification, applying three scenarios for the poverty line. Thereafter, we studied absolute mobility based on a three-way division of the income distribution, using the concepts of vulnerability and middle class status, based on two different scenarios for setting group thresholds. Finally, we analyzed relative mobility based on income quintiles. Synthetic panel analysis makes it possible to draw better-informed conclusions about the nature of economic mobility when compared to a simple cross-sectional examination of the data. For 28 example, we gain insights into whether poverty is more of a chronic or transient nature, and how this changes over time and differs between regions. Moreover, we are able to estimate conditional probabilities, such as chances of escaping poverty. Nevertheless, the caveat remains that these results are not based on true panel data and depend on our willingness to assume that actual income correlation coefficients from other countries can, as a group, proxy for the correlation of Malaysian incomes over time. Considering poverty dynamics, we observe a large decrease in chronic poverty over the study period, to between 2-3 percent of the population over the interval 2019-2022. This decline is evident across all three poverty lines applied. It also means that those living in poverty now have much better chances of escaping poverty. In addition, poverty entry figures have continued to decline. These poverty dynamics went hand-in-hand with larger transfers to low-income households in the wake of the Covid-19 crisis. This is important to bear in mind when interpreting these results. In contrast to the Department of Statistics Malaysia, we do not observe an increase in poverty from 2019 to 2022. This is likely due to a significant increase in the poverty line employed by DOSM from 2019 to 2022. If we pick either of these lines and keep its real value constant over time, we observe a decrease in poverty. Poverty conditions are worst in rural East Malaysia: around 15 percent of the population here lives in chronic poverty, and over each interval in our study, around 10 percent falls into poverty. Those in poverty here have the lowest chance of all Malaysians of escaping poverty. Moreover, those out of poverty face the highest chances of falling into poverty. Looking at three-way income mobility, an important conclusion is that results depend on which income thresholds are used to define the three groups. If we use relatively lower poverty and vulnerability lines, a bit over 50 percent of the population is persistently secure, and downward mobility has declined. However, if we employ higher income thresholds, the persistently secure middle/upper class group is considerably smaller, at slightly more than 30 percent in 2019-2022. Moreover, the share of the population that experiences downward mobility remains sizeable at a bit below 20 percent over the latest transition interval. Furthermore, ethnic differences in mobility are also sizeable under this scenario. Another finding is that we observe increases over time in downward mobility for rural East Malaysia and for the group of East Malaysia Bumiputera. Potentially, such downward mobility happened after having enjoyed upward mobility in the first part of our study period. Notably, an individual in urban Peninsular Malaysia is more than three times more likely to live in economic security than someone in rural East Malaysia 29 Relative mobility, between the quintiles of the income distribution, remained at similar levels across the entire period studied. As is known from other countries, there are sizeable groups who remain stuck in the bottom quintile (around 10 percent of the population) and who keep their position in the top quintile (around 12 percent). Our analysis of mobility patterns in Malaysia points to a number of interesting patterns and distinctions. On the one hand, examining changes in absolute income levels over time, the overriding impression is one of broad and steady improvements in wellbeing. Chronic poverty has fallen over time – a far smaller fraction of the population today appears to experience prolonged episodes of absolute poverty, compared to the early years of our study period. Similarly, there is evidence of it becoming easier for households to lift themselves across the poverty and vulnerability lines over time, and of becoming better able to prevent themselves from falling back into poverty or vulnerability. However, these results do not hold uniformly; certain groups, such as households in rural East Malaysia, have also improved their situation, but are still clearly at a disadvantage. On the other hand, when we consider the capability of Malaysian households to overtake one another – when we examine relative mobility - we see less evidence of changes over time. In relative terms, many of those at the bottom of the income distribution have tended to remain at the lower end, while those at the top have largely stayed at the top. Absolute incomes have risen, but in relative terms mobility seems to be fairly low, and this does not seem to have evolved much over time. The picture that emerges is one of economic growth and development permeating throughout society and lifting incomes across the board, but without dramatically affecting the forces that govern the position of households with respect to one another. 30 References Chaudhuri, Shubham. (2003). “Assessing Vulnerability to Poverty: Concepts, Empirical Methods and Illustrative Examples.” Mimeo. Department of Economics, Columbia University. Chaudhuri, Shubham, Jyotsna Jalan, and Asep Suryahadi. (2002). “Assessing Household Vulnerability to Poverty from Cross-Sectional Data: A Methodology and Estimates from Indonesia.” Columbia University Department of Economics Discussion Papers 0102–52 (April): 36. Colgan, Brian. (2023). “EU-SILC and the potential for synthetic panel estimates”. Empirical Economics 64. pp. 1247–1280. https://doi.org/10.1007/s00181-022-02277-7. Colgan, Brian. (2024). “Long run poverty dynamics in new EU member states”. Mimeo Dang, Hai-Anh, Peter Lanjouw. (2017). “Welfare Dynamics Measurement: Two Definitions of a Vulnerability Line”. Review of Income and Wealth, 63(4): 633-660. Dang, Hai-Anh, Peter Lanjouw. (2023). “Measuring Poverty Dynamics with Synthetic Panels Based on Repeated Cross Sections”. Oxford Bulletin Of Economics And Statistics, 85(3): 599-622. Dang, Hai-Anh, Peter Lanjouw, Jill Luoto, and David McKenzie. (2014). “Using Repeated Cross-Sections to Explore Movements in and out of Poverty”. Journal of Development Economics, 107: 112-128. Deaton, Angus S. and Paxson, Christina H. (1994). "Intertemporal Choice and Inequality." Journal of Political Economy, 102(3): 437-67. Department of Statistics Malaysia (DOSM) (2023). “Poverty in Malaysia”. Putrajaya, Malaysia: Department of Statistics, Malaysia. Krah, K., Montalva Talledo, V., and Tiwari, S. (forthcoming). “Turning Aspirations to Reality: Emerging Middle Class in the East Asia and Pacific Region”. Washington D.C.: World Bank Hérault, Nicolas and Stephen Jenkins. (2019). “How valid are synthetic panel estimates of poverty dynamics?”. Journal of Economic Inequality 17: 51-76. Rongen, G., Ali Ahmad, Z., Lanjouw, P. and Simler, K. (2022). “The Interplay of Ethnic and Regional Inequalities in Malaysian Poverty Dynamics”. Policy Research Working Paper 9898. Washington D.C.: World Bank Rongen, G., Ahmad, Z.A., Lanjouw, P. and Simler, K. (2024). “Regional and ethnic inequalities in Malaysian poverty dynamics”. Journal of Economic Inequality, 22: 101-130. https://doi.org/10.1007/s10888-023-09582-w World Bank (2021). “Staying Afloat”. Malaysia Economic Monitor (December), World Bank, Washington, DC. 31 Appendix Table A Spatial price index 2004 2007 2009 2012 2014 2016 2019 2022 Johor Urban 0.987 0.983 0.978 0.982 0.994 0.996 0.999 0.998 Johor Rural 0.898 0.909 0.904 0.892 0.886 0.881 0.884 0.883 Kedah Urban 0.948 0.942 0.941 0.952 0.947 0.952 0.941 0.931 Kedah Rural 0.883 0.889 0.884 0.887 0.877 0.861 0.851 0.841 Kelantan Urban 0.858 0.875 0.855 0.867 0.880 0.886 0.877 0.876 Kelantan Rural 0.794 0.803 0.800 0.794 0.793 0.780 0.772 0.771 Melaka Urban 0.985 0.977 0.984 0.982 0.985 0.987 0.981 0.968 Melaka Rural 0.906 0.914 0.894 0.884 0.900 0.884 0.878 0.866 N. Sembilan Urban 0.949 0.962 0.943 0.952 0.948 0.953 0.957 0.950 N. Sembilan Rural 0.944 0.964 0.964 0.948 0.924 0.918 0.922 0.915 Pahang Urban 0.976 0.975 0.971 0.977 0.982 0.985 0.969 0.972 Pahang Rural 0.904 0.922 0.915 0.919 0.893 0.882 0.868 0.871 P. Pinang Urban 1.042 1.037 1.042 1.041 1.045 1.050 1.054 1.055 P. Pinang Rural 0.977 0.973 0.977 0.959 0.947 0.944 0.948 0.948 Perak Urban 0.924 0.942 0.937 0.944 0.938 0.944 0.935 0.936 Perak Rural 0.874 0.898 0.888 0.867 0.881 0.851 0.842 0.844 Perlis Urban 0.926 0.919 0.909 0.929 0.921 0.920 0.907 0.903 Perlis Rural 0.862 0.872 0.885 0.868 0.864 0.849 0.837 0.833 Selangor Urban 1.055 1.049 1.038 1.049 1.042 1.049 1.058 1.076 Selangor Rural 0.940 0.941 0.935 0.931 0.921 0.907 0.915 0.931 Terengganu Urban 0.922 0.915 0.919 0.946 0.935 0.946 0.928 0.934 Terengganu Rural 0.919 0.914 0.913 0.899 0.916 0.905 0.888 0.894 Sabah Urban 1.140 1.135 1.151 1.130 1.105 1.076 1.059 1.040 Sabah Rural 1.113 1.074 1.154 1.136 1.096 1.049 1.032 1.013 Sarawak Urban 1.081 1.081 1.090 1.076 1.077 1.073 1.053 1.042 Sarawak Rural 1.049 1.044 1.055 1.019 1.010 0.981 0.963 0.953 WP. KL Urban 1.221 1.204 1.192 1.196 1.190 1.204 1.213 1.213 WP. Labuan Urban 1.140 1.134 1.161 1.081 1.098 1.081 1.050 1.025 WP. Labuan Rural 1.113 1.068 1.184 1.114 1.102 1.062 1.032 1.007 WP Putrajaya Urban 1.161 1.162 1.133 1.098 1.108 1.092 1.162 32 Table B OLS estimates income model. Dependent variable is the logarithm of spatially-adjusted household per capita income Dummies for: 2004 2007 2009 2012 2014 2016 2019 2022 Lower secondary 0.216 0.134 0.182 0.137 0.119 0.115 0.132 0.100 (0.014)*** (0.013)*** (0.013)*** (0.013)*** (0.009)*** (0.011)*** (0.012)*** (0.012)*** Upper secondary 0.530 0.434 0.504 0.434 0.367 0.345 0.364 0.298 (0.014)*** (0.013)*** (0.013)*** (0.013)*** (0.009)*** (0.010)*** (0.011)*** (0.011)*** Vocational 0.762 0.630 0.726 0.657 0.555 0.514 0.531 0.483 (0.029)*** (0.030)*** (0.027)*** (0.025)*** (0.017)*** (0.019)*** (0.017)*** (0.017)*** University (diploma) 1.071 1.020 1.059 0.968 0.785 0.757 0.775 0.686 (0.023)*** (0.021)*** (0.020)*** (0.018)*** (0.012)*** (0.012)*** (0.013)*** (0.014)*** Advanced uni degree 1.452 1.374 1.467 1.361 1.226 1.145 1.153 1.066 (0.025)*** (0.022)*** (0.021)*** (0.017)*** (0.013)*** (0.014)*** (0.014)*** (0.014)*** Ethnicity Chinese 0.562 0.491 0.437 0.476 0.400 0.406 0.439 0.389 (0.014)*** (0.012)*** (0.012)*** (0.011)*** (0.008)*** (0.009)*** (0.009)*** (0.009)*** Ethnicity Indian 0.210 0.145 0.070 0.102 0.090 0.097 0.124 0.110 (0.020)*** (0.019)*** (0.017)*** (0.018)*** (0.011)*** (0.012)*** (0.013)*** (0.013)*** Ethnicity Other 0.173 0.070 0.089 0.132 0.129 -0.037 0.001 0.135 (0.065)*** (0.089) (0.071) (0.056)** (0.050)** (0.024) (0.027) (0.044)*** Female-headed 0.073 0.007 0.000 -0.028 -0.029 -0.043 -0.036 -0.026 (0.016)*** (0.015) (0.013) (0.012)** (0.008)*** (0.008)*** (0.008)*** (0.007)*** Rural -0.287 -0.238 -0.254 -0.221 -0.197 -0.184 -0.157 -0.196 (0.012)*** (0.011)*** (0.012)*** (0.011)*** (0.008)*** (0.009)*** (0.009)*** (0.008)*** Peninsular 0.307 0.248 0.286 0.232 0.210 0.263 0.268 0.249 (0.015)*** (0.014)*** (0.015)*** (0.013)*** (0.009)*** (0.008)*** (0.008)*** (0.008)*** b_yr1940 0.478 (0.029)*** b_yr1945 0.376 0.417 0.420 (0.019)*** (0.021)*** (0.030)*** b_yr1950 0.333 0.374 0.378 0.323 0.346 (0.018)*** (0.017)*** (0.017)*** (0.019)*** (0.019)*** b_yr1955 0.148 0.207 0.288 0.310 0.300 0.261 0.120 (0.016)*** (0.015)*** (0.015)*** (0.015)*** (0.010)*** (0.012)*** (0.021)*** b_yr1960 -0.003 0.082 0.086 0.152 0.184 0.177 0.148 0.084 (0.015) (0.015)*** (0.015)*** (0.014)*** (0.009)*** (0.010)*** (0.011)*** (0.012)*** b_yr1970 0.119 0.092 0.043 -0.028 -0.038 -0.084 -0.133 -0.128 (0.018)*** (0.016)*** (0.015)*** (0.013)** (0.009)*** (0.010)*** (0.010)*** (0.010)*** b_yr1975 0.286 0.210 0.117 0.012 -0.012 -0.101 -0.196 -0.245 (0.022)*** (0.017)*** (0.016)*** (0.014) (0.009) (0.010)*** (0.010)*** (0.010)*** b_yr1980 0.270 0.239 0.133 0.043 -0.072 -0.194 -0.282 (0.026)*** (0.018)*** (0.015)*** (0.010)*** (0.010)*** (0.010)*** (0.010)*** b_yr1985 0.141 0.048 -0.043 -0.161 -0.257 (0.022)*** (0.011)*** (0.011)*** (0.011)*** (0.010)*** b_yr1990 -0.100 -0.152 -0.216 (0.019)*** (0.013)*** (0.011)*** 33 b_yr1995 -0.244 (0.015)*** Constant 5.644 5.803 5.774 6.061 6.307 6.438 6.571 6.790 (0.021)*** (0.020)*** (0.021)*** (0.019)*** (0.013)*** (0.013)*** (0.013)*** (0.014)*** R2 0.438 0.411 0.425 0.416 0.391 0.383 0.381 0.391 N 29,065 29,918 34,752 35,222 67,514 63,447 65,367 68,033 * p<0.1; ** p<0.05; *** p<0.01 Birth cohort dummies cover five-year bins. Cohort 1965-1969 is the omitted dummy. Sample is limited to synthetic panel age range. 34 Table C Population-level poverty transitions Synthetic panel transition tables Estimates are for the share of the population in a particular transition outcome. The poverty line used is RM 527. (y1==2004&y2==2007) LB MIN_2SD POINT PLUS_2SD UB N Poor to poor 0.202 0.229 0.253 0.277 0.355 29917 Poor to non-poor 0.035 0.111 0.136 0.16 0.188 29917 Non-poor to poor 0.004 0.079 0.103 0.128 0.157 29917 Non-poor to non-poor 0.453 0.484 0.508 0.532 0.606 29917 (y1==2007&y2==2009) LB MIN_2SD POINT PLUS_2SD UB N Poor to poor 0.175 0.207 0.232 0.257 0.333 34907 Poor to non-poor 0.016 0.09 0.115 0.14 0.174 34907 Non-poor to poor 0.006 0.082 0.107 0.131 0.164 34907 Non-poor to non-poor 0.487 0.522 0.547 0.571 0.645 34907 (y1==2009&y2==2012) LB MIN_2SD POINT PLUS_2SD UB N Poor to poor 0.124 0.137 0.157 0.176 0.241 35682 Poor to non-poor 0.084 0.149 0.168 0.187 0.203 35682 Non-poor to poor 0 0.061 0.08 0.1 0.117 35682 Non-poor to non-poor 0.556 0.576 0.595 0.615 0.675 35682 (y1==2012&y2==2014) LB MIN_2SD POINT PLUS_2SD UB N Poor to poor 0.056 0.076 0.091 0.106 0.14 67986 Poor to non-poor 0.071 0.108 0.123 0.138 0.157 67986 Non-poor to poor 0 0.038 0.053 0.069 0.084 67986 Non-poor to non-poor 0.703 0.717 0.733 0.748 0.789 67986 (y1==2014&y2==2016) LB MIN_2SD POINT PLUS_2SD UB N Poor to poor 0.03 0.046 0.057 0.069 0.104 63994 Poor to non-poor 0.03 0.069 0.08 0.092 0.104 63994 Non-poor to poor 0.001 0.039 0.05 0.062 0.074 63994 Non-poor to non-poor 0.792 0.8 0.812 0.824 0.866 63994 (y1==2016&y2==2019) LB MIN_2SD POINT PLUS_2SD UB N Poor to poor 0.021 0.03 0.039 0.048 0.085 66422 Poor to non-poor 0.02 0.057 0.066 0.075 0.081 66422 Non-poor to poor 0 0.037 0.046 0.055 0.064 66422 Non-poor to non-poor 0.834 0.84 0.849 0.858 0.895 66422 (y1==2019&y2==2022) LB MIN_2SD POINT PLUS_2SD UB N Poor to poor 0.014 0.02 0.027 0.034 0.06 69545 Poor to non-poor 0.022 0.05 0.057 0.064 0.068 69545 Non-poor to poor 0 0.029 0.036 0.043 0.047 69545 Non-poor to non-poor 0.872 0.873 0.88 0.887 0.917 69545 35 Table D Relative mobility aggregate transition matrices Synthetic panel transition tables - relative mobility Estimates are for the percentage of the population in a particular transition outcome. Rows give the quintile in year 1, columns the quintile in year 2. Each set of two rows contains the 2 SD band around the point estimate (plus/minus). (y1==2004&y2==2007) Q1 Q2 Q3 Q4 Q5 Q1 8.9 4.9 1.6 0.2 0 - 12.3 5.2 3.1 1.5 0.3 Q2 5.2 5.2 4.5 2.1 0.2 - 5.4 7.1 5 3.3 1.2 Q3 1.8 4.6 4.8 4.5 1.2 - 3.3 5.3 6.5 5 2.6 Q4 0.3 2.5 4.7 5.5 4.5 - 1.7 3.5 5.4 7.6 4.9 Q5 0 0.3 1.6 5.3 10.9 - 0.4 1.5 3 5.4 14.1 (y1==2007&y2==2009) Q1 Q2 Q3 Q4 Q5 Q1 9.2 5 1.4 0.1 0 - 12.9 5.4 3.1 1.3 0.3 Q2 5 5.5 4.7 1.8 0.1 - 5.3 7.7 5.2 3.1 1.2 Q3 1.5 4.9 5.1 4.4 1.1 - 3.2 5.5 7.1 4.9 2.6 Q4 0.1 2.2 4.9 5.3 4.6 - 1.5 3.5 5.7 7.6 5 Q5 0 0.1 1.4 4.8 11 - 0.3 1.3 2.9 5 14.2 (y1==2009&y2==2012) Q1 Q2 Q3 Q4 Q5 Q1 7.8 5 2 0.5 0 - 11 5.1 3.2 1.8 0.5 Q2 5.4 5.4 4.5 2.6 0.5 - 5.6 6.8 4.9 3.5 1.6 Q3 2.4 4.8 4.8 4.5 1.9 - 3.7 5.4 6 5 2.9 Q4 0.6 2.8 4.3 4.9 4.6 - 2 3.6 4.9 6.4 4.7 Q5 0 0.6 2.1 5.2 10.1 - 0.7 1.8 3.1 5.3 13 36 (y1==2012&y2==2014) Q1 Q2 Q3 Q4 Q5 Q1 8.6 4.2 1.4 0.2 0 - 11.9 4.6 2.8 1.4 0.4 Q2 5.8 5.2 4.4 2.1 0.2 - 5.9 7 4.7 3.2 1.3 Q3 2.2 4.6 4.6 4.2 1.4 - 3.6 5.3 6.1 4.7 2.7 Q4 0.5 2.7 4.6 5.2 4.7 - 2 3.6 5.4 7.2 5.1 Q5 0 0.4 1.8 5.3 11.5 - 0.5 1.6 3.1 5.4 14.7 (y1==2014&y2==2016) Q1 Q2 Q3 Q4 Q5 Q1 9.3 4.9 1.7 0.3 0 - 12.8 5.1 3.2 1.6 0.4 Q2 5.2 5 4.4 2.1 0.2 - 5.4 6.7 4.9 3.1 1.4 Q3 1.9 4.4 4.7 4.2 1.4 - 3.4 5.1 6.2 4.7 2.7 Q4 0.4 2.5 4.5 5 4.6 - 1.8 3.4 5.3 6.9 4.9 Q5 0 0.3 1.9 5.4 11.3 - 0.5 1.6 3.2 5.5 14.6 (y1==2016&y2==2019) Q1 Q2 Q3 Q4 Q5 Q1 9 5.2 2 0.5 0 - 12.3 5.3 3.3 1.8 0.5 Q2 5.1 5.1 4.4 2.4 0.4 - 5.2 6.6 4.9 3.2 1.6 Q3 2.1 4.6 4.7 4.3 1.8 - 3.4 5.2 6.1 4.8 3 Q4 0.5 2.6 4.4 4.9 4.8 - 1.8 3.4 5 6.5 5 Q5 0 0.5 2 5.2 10.5 - 0.6 1.7 3.2 5.2 13.6 (y1==2019&y2==2022) Q1 Q2 Q3 Q4 Q5 Q1 8.8 5.2 2 0.5 0 - 12 5.3 3.3 1.7 0.5 Q2 5.2 5.1 4.4 2.4 0.4 - 5.3 6.6 4.8 3.2 1.6 Q3 2.1 4.5 4.6 4.2 1.8 - 3.4 5.1 5.9 4.7 3 Q4 0.5 2.7 4.3 4.9 4.9 - 1.8 3.4 5 6.4 5 Q5 0 0.5 2.1 5.3 10.6 - 0.6 1.8 3.3 5.4 13.6 37 Figure 16 Poverty dynamics 2004 - 2022 (alternative poverty line – RM 618) Figure 17 Income distribution and group thresholds 38 Figure 18 Mobility by ethnic group - scenario 2 39 Figure 19 Mobility by geography - scenario 2 40