ADBI Working Paper Series

THE FATE OF JOB CREATION

IN THE PHILIPPINES AMID THE
AUTOMATION REVOLUTION:
A FIRM-LEVEL ANALYSIS

Raymond Gaspar
and Nicholas Harris

No. 1081
February 2020

Asian Development Bank Institute

 Raymond Gaspar is a consultant at the Economic Research and Regional Cooperation
Department of the Asian Development Bank. Nicholas Harris is a senior lecturer at the
Australian College of Applied Psychology.
The views expressed in this paper are the views of the author and do not necessarily reflect
the views or policies of ADBI, ADB, its Board of Directors, or the governments they
represent. ADBI does not guarantee the accuracy of the data included in this paper and
accepts no responsibility for any consequences of their use. Terminology used may not
necessarily be consistent with ADB official terms.
Working papers are subject to formal revision and correction before they are finalized and
considered published.

The Working Paper series is a continuation of the formerly named Discussion Paper series;
the numbering of the papers continued without interruption or change. ADBI’s working papers
reflect initial ideas on a topic and are posted online for discussion. Some working papers may
develop into other forms of publication.

Suggested citation:

Gaspar, R. and N. Harris. 2020. The Fate of Job Creation in the Philippines Amid the
Automation Revolution: A Firm-Level Analysis. ADBI Working Paper 1081. Tokyo: Asian
Development Bank Institute. Available: https://www.adb.org/publications/fate-job-creation-
philippines-amid-automation-revolution

Please contact the authors for information about this paper.

Email: [email protected], [email protected]

The authors thank the participants in the Conference on Human Capital Development for
Inclusive Growth and Shared Prosperity held on 18–19 July 2018 in Chengdu, People’s
Republic of China, for their valuable comments and suggestions.

Asian Development Bank Institute

Kasumigaseki Building, 8th Floor
3-2-5 Kasumigaseki, Chiyoda-ku
Tokyo 100-6008, Japan

Tel: +81-3-3593-5500
Fax: +81-3-3593-5571
URL: www.adbi.org
E-mail: [email protected]

© 2020 Asian Development Bank Institute

ADBI Working Paper 1081 Gaspar and Harris

Abstract

Using the World Bank’s Enterprise Survey for the Philippines, our analysis found that, on
average, firms that deployed partial or full automation in their operations were more likely
to hire more workers than fewer relative to their counterpart firms. These results suggest a net
positive employment impact of automation, at least in the short run. The findings differ,
however, by industries—net job creation is significant only among manufacturing firms. In
contrast, we found that firms in the service industry have a higher likelihood of net job losses
as they adopt partial or full automation. Indeed, there is an expectation that firms’ decision to
automate will cause labor market disruptions, which deserve immediate policy interventions.
Human capital development, such as education and training, would be a critical policy lever to
prevent the adverse effect of job displacement on the economic, social, and psychological
well-being of individuals. The government should prepare workers to acquire the right set of
skills and competencies, making them more flexible and able to fulfill new tasks amid further
adoption of automation. Otherwise, the anxiety over the future loss of work will become
a reality.

Keywords: automation, labor market disruptions, employment, Philippines

JEL Classification: J21, J23, J24

ADBI Working Paper 1081 Gaspar and Harris

Contents

1. INTRODUCTION ......................................................................................................... 1

2. THE STATE OF WORK AND AUTOMATION IN THE PHILIPPINES ......................... 2

3. THEORETICAL FRAMEWORK................................................................................... 4

4. EMPIRICAL ANALYSIS .............................................................................................. 5

4.1 Firms’ Hiring Decisions .................................................................................... 5

4.2 Firms’ Decision and Propensity to Automate................................................... 6
4.3 Estimation Strategy ......................................................................................... 6
4.4 Analyzing the Disproportionate Effect on Female and Low-Skilled Workers... 7

5. EMPIRICAL FINDINGS ............................................................................................... 8

6. CONCLUSION AND POLICY IMPLICATIONS ......................................................... 15

REFERENCES ..................................................................................................................... 17

APPENDIX ............................................................................................................................ 18
ADBI Working Paper 1081 Gaspar and Harris

1. INTRODUCTION
Technological advances create both economic opportunities and challenges. One such
challenge is the anxiety that arises over the future of work (Innes and Morrison 2017).
Productivity improvement from the use of computers and the deployment of robots may
reduce the demand for labor to produce a good or deliver a service. The risk of
automation, however, seems to pose a greater concern for jobs involving tasks that are
repetitive, manual, and simple, which are more automatable than more diverse,
cognitive, emotional, and complex tasks (Asian Development Bank 2018).
The literature on the impact of automation on labor has offered mixed views. Autor (2015)
argued that, while automation generally intends to substitute manual labor, the demand
for human workers to perform other tasks would also increase owing to machines
producing output more quickly and cheaply. Bessen (2015) followed the same line of
thought, noting the historical example of the automation of weaving in the US in the 19th
century. The demand for weavers declined because of the increased productivity, but
the downward adjustment of the price of cloth pushed the demand up, which in turn led
to a greater demand for weavers. Others, however, have argued that disruptions from
more advanced technologies, such as artificial intelligence, deep learning, the Internet
of Things, and 3D printing, are likely to be wider in scope than the previous industrial
revolutions. Ford (2015) explained that the broad-based adoption of technologies, such
as computers and e-commerce, across all industries could be an indication of how widely
the global corporate world could adopt new technologies, such as automation. This view
was shared by Schwab (2016), noting the expectation that the disruptive effects of
advanced technologies will be wide in scope and on a scale large enough to displace
jobs.
The potential labor market implications of automation also apply in the context of
developing economies while they remain at the initial stages of technological adoption.
The World Bank (2016) estimated that a large proportion of occupations in developing
economies, such as the People’s Republic of China, India, and Ethiopia, will become
automated. However, the possible displacement of these jobs will remain at bay due
to the moderate pace of technological adoption and low labor costs in these areas. In
Southeast Asia, labor-intensive jobs in electronics, automotive, textiles, and retail
services are particularly at risk from automation (Chang, Rynhart, and Huynh 2016).
Further, given their capacity to invest heavily in automation and artificial intelligence,
multinational firms, often from advanced economies, may decide to relocate their
production facilities to their home country or where their market is located, displacing
workers from developing economies.
This paper narrows the scope and considers the case of the Philippines labor market.
The country holds firmly onto employment creation to achieve the inclusive growth
goal that the Philippine Development Plan 2017–2022 stipulated. However, the risk of
automation, which could infiltrate the Philippines labor market just like any other labor
market in the region, could equate to massive layoffs and potentially increase
unemployment. While the country’s manufacturing sector has been slow to adopt
automation, with only three installed industrial robots for every 10,000 employees in 2016
(International Federation of Robotics 2018), it is apparent that the number of business
process outsourcing firms deploying robotic process automation is growing amid
observed productivity gains. The empirical analysis of the paper aims to evaluate the
labor market disruptions of firm-level automation decisions and provide policy insights
not only to manage potential adverse consequences but also to take advantage of some
opportunities that will arise in the Philippines labor market.

1
ADBI Working Paper 1081 Gaspar and Harris

This paper also contributes to the literature and the ongoing debate among academics,
scholars, and policy makers on how further advances in technologies will disrupt the
labor market. Much of the literature on this topic has dwelled heavily on estimating jobs
at risk of automation by analyzing the automatability of the tasks involved, the pace and
extent of adoption, and the ease of labor substitution, among others. An innovation of
this paper lies in its evaluation of the issue through the lens of enterprises, which we
believe could capture the interplay of the relevant labor demand and supply factors,
including firms’ technological capacity, the task composition of jobs, and workers’ skills,
among others.
Using the World Bank’s Enterprise Survey for the Philippines, this paper attempts to
understand generally how automation decisions change the hiring behavior of firms
operating in the Philippines. The paper aims to determine whether firms that are more
inclined to adopt new and advanced technologies are more likely to reduce human labor,
particularly in low-skilled positions. Broadly, the results suggest that firms that have
adopted partial or full automation to develop their new products or services, on average,
are likely to hire more workers than their counterpart firms. Firms’ decision to automate
is associated with significant growth in the hiring of new workers, a scenario, based on
the estimates, that is much more likely to occur than massive worker layoffs.
Interestingly, the study finds that the greater propensity of firms to automate is associated
with a larger proportion of the total full-time employment of unskilled workers. Despite
the observations, these firms offer more formal training to their workers. The potential
rise in the demand for unskilled workers who can handle non-routine manual tasks could
explain such a finding. Last, the paper determines how increasing automation could
benefit the female workforce in the Philippines.
The next section provides a general overview of the Philippine labor market as well as
the pace of the automation revolution in the country. Section 3 explains the conceptual
framework and the empirical approach that the analysis in this paper follows. Section 4
presents the empirical findings. Lastly, section 5 concludes and recommends policy
directions concerning how the Philippines could take advantage of the economic
opportunities and face the challenges that the automation revolution presents.

2. THE STATE OF WORK AND AUTOMATION

IN THE PHILIPPINES
Will the automation revolution worsen the job prospects of the Filipino workforce? To
answer this question, first we needed to examine the labor market structure of the
Philippines as well as the propensity and pace of automation of firms operating in
the Philippines.
Deriving from the January 2018 round of the Philippine Labor Force Survey that the
Philippine Statistics Authority (2018) conducted, the Philippine labor force climbed nearly
5% to 44 million from 42 million during the same period in 2017. Of the total labor force,
only 38% is female, and the participation rate for women remains low relative to that of
their male counterparts. The number of employed persons increased even more, by
6.1%, causing the unemployment rate to drop to 5.3%. It is worth noting that male and
female employment grew at a similar pace. Many of the gains in total employment,
however, are attributable to the larger amount of part-time employment or those with
fewer than 40 working hours per week; thus, underemployment remains a concern that
the country should address.

2
ADBI Working Paper 1081 Gaspar and Harris

By industry, the service sector remains the major source of employment in the country,
more than a third of which is involved in retail and wholesale trade. The agriculture sector
still accounts for more than a quarter of the total employed persons, despite its falling
share of the total gross value added. The manufacturing and construction sectors also
employ a large workforce, with their individual share of the total employment exceeding
8%. Another important employment contributor in the country is the business process
outsourcing sector; estimates have indicated that it employs about 1 million workers.
Based solely on the sectoral decomposition of employment, without considering the pace
of automation adoption in these industries, the disruptions in the labor market in the
Philippines could be massive. According to the McKinsey Global Institute (2017), the
technical potential for automation of the sectors that generate most of the employment
in the Philippines is considerably high.
We also intended to take a close look at the occupational make-up of the country’s labor
market. The January 2018 labor force survey showed that 27.6% of Filipino workers are
involved in elementary occupations and perform simple and routine tasks using handheld
tools and considerable physical effort. It is evident that the majority
of the employment gains come from this occupation group. It is followed by those
performing managerial roles (16.3%). There are also many service and sales workers
(14.7%), skilled agricultural, fishery, and forestry workers (13.1%), craft and related trade
workers (7.1%), and plant and machine operators and assemblers (6.4%). These
occupations perform tasks that are easily automatable, classified as manual and routine.
The call center jobs in the country also fall into this category despite the involvement of
some cognitive tasks. Meanwhile, professionals and technicians performing technical
and related tasks comprise only 9% of the total employment.
Aside from assessing the feasibility of partial and full automation of tasks and their
application in broad sets of industries to determine how vulnerable the country’s
labor market will be in the age of automation, it is equally worthwhile to examine the issue
from the perspective of enterprises. While there are tasks that are easily automatable,
firms also consider the costs associated with upgraded or completely automated
production. The economic feasibility of automation, having considered all the costs, such
as capital, time, and effort, will influence and shape firms’ decision regarding the adoption
of automation (McKinsey Global Institute 2017).
Using the 2016 International Labour Office survey, Orozco (2017) found a low technology
uptake among 575 surveyed enterprises in the Philippines. Of the total respondent firms,
only 27% had upgraded their technology and an even lower 19% had spent money on
research and development activities. When asked about the key reason for the low
adoption, more than 30% of the respondents cited the high fixed capital cost involved in
completing such activities. Compared with other surveyed Southeast Asian countries,
such as Cambodia, Indonesia, Thailand, and Viet Nam, a smaller number of firms in the
Philippines mentioned the lack of skilled operators
as a major barrier to technology upgrading. Despite the low uptake, many firms in the
Philippines perceive that technological advances would bring huge economic
opportunities to their enterprise by 2025. This may reflect the enthusiasm of firms about
deploying automation technologies soon that will have labor market implications in
the country.
The labor impact of the automation revolution remains generally uncertain, but what is
certain is that the Philippine labor market is a pool of medium- and high-skilled workers,
which offers the opportunity and flexibility to face challenges once full-scale automation
has taken place in the near future.

3
ADBI Working Paper 1081 Gaspar and Harris

3. THEORETICAL FRAMEWORK
This study applied the theory of two-factor labor demand to understand firms’ hiring
decisions given their propensity to adopt automated processes. The framework guided
us to derive an empirically tractable specification that could determine whether and to
what extent automation affects the hiring decisions of firms operating in the Philippines.
Consider a firm with a production function, 𝑌𝑌 = 𝐴𝐴𝐴𝐴(𝐿𝐿, 𝐾𝐾), and assume the conditions 𝐹𝐹 ′ >
0 , 𝐹𝐹 ′′ < 0 , and constant returns to scale. Firms maximize their profit,
𝜋𝜋 = 𝐹𝐹 (𝐿𝐿, 𝐾𝐾 ) − 𝑤𝑤𝑤𝑤 − 𝑟𝑟𝑟𝑟, where 𝑤𝑤 refers to the cost of labor 𝐿𝐿 and 𝑟𝑟 to the cost of capital
𝐾𝐾, by choosing the right mix of factor inputs. Taking the ratio of the first-order conditions,
𝐹𝐹𝐿𝐿 − 𝜆𝜆𝜆𝜆 = 0 and 𝐹𝐹𝐾𝐾 − 𝜆𝜆𝜆𝜆 = 0, we obtained the profit-maximizing condition in which the
marginal rate of technical substitution, 𝐹𝐹𝐿𝐿 /𝐹𝐹𝐾𝐾 , equals the relative price of labor or the
factor–price ratio, 𝑤𝑤/𝑟𝑟. In short, the employment decisions of firms depend on how they
respond to changes in the relative prices of factor inputs.
Within firms, using the above framework, there will be three interrelated scenarios in
which it is possible to determine how automation disrupts the labor market. These
scenarios depend on different forces and factors that could shape the automation
decisions of firms.
Scenario 1. Automation leads to job displacement.
𝑑𝑑ln(𝐾𝐾 ⁄𝐿𝐿 )
This will be the case when the elasticity of substitution, 𝜎𝜎 = , derived from the
𝑑𝑑ln(𝑤𝑤 ⁄𝑟𝑟)
above framework, holding the output constant, approaches infinity, which implies that the
factor inputs are perfect substitutes. This scenario will depend on how easily firms can
automate activities that human labor previously performed. First, firms are most likely to
consider the amount of fixed capital costs. Automation displaces human labor if the
marginal cost associated with it tends to be less than a worker’s wages. Over time, one
would expect the marginal cost of automation to be lower than the wages, which may
lead firms to decide to substitute human labor with the automated system. Second, firms
might consider the social acceptance of technology. This scenario also depends on the
work dimension, that is, whether tasks are routine and whether they require cognitive
skills. Jobs that machines are more likely to replace involve mainly routine and manual
tasks.
Scenario 2. Automation increases productivity, thereby increasing the demand for other
tasks for human labor.
The labor substitution mechanism in the first scenario would certainly be one of the major
forces that would cause labor disruptions within firms. Considering the first scenario,
Scenario 2 asserts that the potential increase in productivity owing to automation
adoption will lead to a higher demand for labor. Productivity improvement lowers
production costs, which will lead to lower prices of goods and services. As
the law of demand and supply suggests, it will result in a higher demand for goods
produced and services delivered, which will subsequently encourage firms to hire more
laborers, especially to handle the non-automated tasks involved in the production.
Scenario 3. Automation produces new tasks for human labor.

4
ADBI Working Paper 1081 Gaspar and Harris

Closely related to the second scenario, Scenario 3 exemplifies the case in which
automation creates new tasks that require human labor. Acemoglu and Restrepo (2018)
called it the reinstatement effect, which could arise when, following the automation of
some or most of the production process, there are new functions
and activities that only humans can perform or that humans can perform better
than machines. For example, maintenance functions or other technical skills like
programming will be necessary to ensure that machines function well.

4. EMPIRICAL ANALYSIS
We used the World Bank Enterprise Survey dataset for the Philippines for the year 2015
in conducting the empirical exercises. The survey used the stratified random sampling
method, which resulted in a sample of 1,335 firms across different industries,
establishment sizes, and locations (Table 1).

Table 1: Respondent Firms by Firm Size, Industry, and Region

Region
Metro Central Metro
Firm Size and Industry Manila Luzon Calabarzon Cebu Total
Small (5–19 employees)
Manufacturing 128 67 80 58 333
Retail trade 37 14 18 10 79
Other services 30 10 12 10 62
Medium (20–99 employees)
Manufacturing 147 82 108 48 385
Retail trade 22 6 9 1 38
Other services 26 7 11 8 52
Large (100 and more employees)
Manufacturing 93 49 107 54 303
Retail trade 18 10 11 7 46
Other services 17 8 5 7 37
Total 518 253 361 203 1,335
Source: World Bank (2015).

4.1 Firms’ Hiring Decisions

We derived the dependent variable, the hiring choices of firms, using the employment
information that the survey questionnaire sought. The respondents reported the number
of permanent and full-time employees during the last fiscal year (𝐹𝐹𝐹𝐹𝐸𝐸0 ) and three fiscal
years ago (𝐹𝐹𝐹𝐹𝐸𝐸−3 ). Given the information on employment for the two different periods,
firms’ hiring decision 𝑦𝑦𝑛𝑛𝑛𝑛 can take the following categorical values:

1 if 𝐹𝐹𝐹𝐹𝐸𝐸0 < 𝐹𝐹𝐹𝐹𝐸𝐸−3

𝑦𝑦𝑛𝑛𝑛𝑛 = �2 if 𝐹𝐹𝐹𝐹𝐸𝐸0 = 𝐹𝐹𝐹𝐹𝐸𝐸−3
3 if 𝐹𝐹𝐹𝐹𝐸𝐸0 > 𝐹𝐹𝐹𝐹𝐸𝐸−3

5
ADBI Working Paper 1081 Gaspar and Harris

To observe the depth of labor market disruptions, we calculated the compounded annual
growth of the permanent and full-time employment during the three-year period
(𝐹𝐹𝐹𝐹𝐸𝐸𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶 ) and broke down categories 1 and 3, classifying the double-digit compounded
annual growth rate (both positive and negative) as indicating high labor market
disruption, with single-digit growth indicating low expansion or reduction. Hence, we
derived the following hiring decision choices:

1 if 𝐹𝐹𝐹𝐹𝐸𝐸𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶 ≤ −10%
⎧2 if – 10% < 𝐹𝐹𝐹𝐹𝐸𝐸𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶 < 0
⎪
𝑦𝑦𝑛𝑛𝑛𝑛 = 3 if 𝐹𝐹𝐹𝐹𝐸𝐸𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶 = 0
⎨4 if 0 < 𝐹𝐹𝐹𝐹𝐸𝐸𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶 < 10%
⎪
⎩5 if 𝐹𝐹𝐹𝐹𝐸𝐸𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶 ≥ 10%

4.2 Firms’ Decision and Propensity to Automate

We denoted the automation decision with a dummy variable that takes the value of
one if a firm had automated its manual processes either fully or partially to introduce a
new and improved product or service and zero otherwise. To check the robustness
of the results, we derived an index measuring firms’ propensity to automate with
the following components: (i) the adoption of new technology, (ii) the undertaking
of research and development activities, (iii) the use of foreign-licensed technology,
(iv) the existence of a company website, and (v) purchases or licensing of any patented
or non-patented inventions or other types of knowledge to develop new products
and services. For simplicity, the index imposes equal weights on each component
and overall, taking values ranging between zero (less likely to automate) and one (more
likely to automate).

4.3 Estimation Strategy

Guided by the conceptual framework, we derived an empirically tractable specification
that attempts to model the hiring decisions of firms operating in the Philippines and
determine how their automation decisions shape them. Firms are likely to choose an
alternative that maximizes their profit. While we cannot easily predict a firm’s choice, we
can derive the probability that firm 𝑛𝑛 will choose 𝑖𝑖 among the alternatives by maximizing
the log-likelihood function over parameters 𝛽𝛽 of observed explanatory variables 𝑥𝑥:
′
𝐽𝐽 𝑒𝑒 𝑥𝑥𝑛𝑛 𝛽𝛽𝑖𝑖
ln𝐿𝐿(𝛽𝛽 ) = ∑𝑁𝑁
𝑖𝑖=1 ∑𝑗𝑗=1 𝑦𝑦𝑛𝑛𝑛𝑛 ln 𝑥𝑥 𝛽𝛽′ (1)
∑𝑗𝑗 𝑒𝑒 𝑛𝑛 𝑗𝑗

where 𝑦𝑦𝑛𝑛𝑛𝑛 denotes the hiring choice of firm 𝑛𝑛 with 𝑗𝑗 as the base category over the
five choices. Since we are interested in understanding how firms’ automation decision
influences their hiring decision, among the explanatory variables, we focus on the
variable denoting automation, specifically 𝑥𝑥𝑘𝑘 .

4.3.1 Control Variables

The explanatory variables controlling for firm-specific characteristics and sector and
location effects include labor productivity (measured as sales divided by the total number
of employees and expressed in logs), the age of the firm (which enters
in a quadratic form), the size of the firm (small, medium, and large), the sectoral
classification (manufacturing, retail trade services, and other services), and the regional
location (Metro Manila, Central Luzon, Calabarzon, and Metro Cebu).

6
ADBI Working Paper 1081 Gaspar and Harris

4.3.2 Robustness Check

To check whether the results are robust, we conducted similar multinomial logistic
regressions but used the calculated index of automation propensity in lieu of the
automation dummy from the baseline model. This alternative indicator is also an attempt
to address the potential time mismatch involved in the automation dummy variable and
firms’ hiring choices. The mismatch arises from not knowing exactly
when firms adopted automation during the past three fiscal years, which could distort the
association between automation and hiring decisions. The index of automation
propensity, on the other hand, acts like an additional observed firm-specific characteristic
that may influence the depth of employment changes within firms.

4.4 Analyzing the Disproportionate Effect on Female

and Low-Skilled Workers
To contribute to the ongoing policy debate on the future of work amid automation,
we assessed how firms’ propensity to automate disrupts the labor market structure. Is
there potential bias against women workers, or will automation perpetuate social
inequality by favoring skilled workers over unskilled ones? To answer these questions,
we estimated ordinary least squares regressions of the following specification:

𝑦𝑦𝑖𝑖 = 𝛼𝛼 + 𝛽𝛽𝛽𝛽𝛽𝛽𝛽𝛽𝛽𝛽𝛽𝛽𝛽𝛽𝛽𝛽𝛽𝛽𝛽𝛽𝑛𝑛𝑖𝑖 + 𝐱𝐱 ′ 𝛾𝛾 + 𝜀𝜀𝑖𝑖 , (2)

where 𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑚𝑚𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑛𝑛𝑖𝑖 refers to firm 𝑖𝑖’s derived index of automation propensity and 𝐱𝐱 ′ is a
vector of control variables that include labor productivity, the age of the firm, the size of
the firm, the sector to which the firm belongs, and the firm’s location. 𝜀𝜀𝑖𝑖 denotes the
stochastic error term.
The dependent variable 𝑦𝑦𝑖𝑖 represents the derived ratios to total full-time employment
of the following: (i) female employment, (ii) non-production employment, (iii) female non-
production employment, (iv) unskilled employment, (v) female unskilled employment,
and (vi) female temporary or seasonal employment. Here, we tried to determine whether
automation tends to be skill biased, which favors skilled over unskilled workers. We also
attempted to check whether automation tends to favor male over female workers,
particularly for unskilled, non-production, and temporary and seasonal workers.
Table 2 displays the descriptive statistics of the variables that the study used. The firms
under study are heterogeneous in terms of age, with a mean of 22 years but a standard
deviation of 15 years. The labor productivity (in logs) across firms exhibits little
dispersion, while the opposite is apparent for automation adoption and the index of
automation propensity. As we expected, given the three-year period of the survey that
the study used, the hiring decision variable nears the value of three, which suggests no
employment reduction or expansion, as we defined above.

7
ADBI Working Paper 1081 Gaspar and Harris

Table 2: Summary Statistics of the Variables in the Study

Std
Variables Obs. Mean Dev. Min. Max.
Firm’s age (difference between 2015 and the year 1,311 21.998 15.551 1 161
of establishment)
Labor productivity (expressed in logs) 1,209 14.004 1.782 7.706 21.115
Automation adoption (takes the value 1 if yes; otherwise 0) 1,319 0.366 0.482 0 1
Index of automation propensity
i. Adoption of foreign-licensed technology 1,335 0 1
ii. Use of new technologies 1,335 0 1
iii. Undertaking of research and development activities 1,335 0 1
iv. Existence of a company website 1,335 0 1
v. Purchases or licensing of inventions, patented or not 1,335 0 1
Overall index 1,335 0.240 0.229 0 1
Hiring decision 1,219 3.185 0.957 1 5
Derived ratios
Female to total full-time employment 1,325 0.087 0.198 0 1
Non-production to total full-time employment 1,004 0.325 0.203 0 1
Female non-production to total full-time employment 1,004 0.000 0.004 0 0.082
Unskilled to total full-time employment 998 0.118 0.204 0 0.883
Female unskilled to total full-time employment 998 0.040 0.106 0 0.706
Female temporary to full-time employment 1,300 0.189 0.323 0 1

Source: Authors’ calculation from World Bank (2015).

5. EMPIRICAL FINDINGS
The initial assessment of the Philippine labor market structure in Section 2 revealed the
potential vulnerability of the country in the event of wide-scale automation adoption.
Given the current low uptake of new technologies among Philippine enterprises, we
expected few or no significant labor market disruptions to take place in the near term.
However, for policy purposes, it is essential to understand as early as possible how firms’
decisions or intention to adopt either partially or fully automated processes shape their
hiring decisions.
Table 3 presents the number of firms that did and did not adopt partially or fully
automated processes to develop new products or services during the last three fiscal
years. One can easily observe that, consistent with the ILO survey results, relatively few
(i.e., only 37% of the respondent firms) adopted automation to the partial or full extent.
The table also presents their corresponding hiring decisions. It is worth noting that firms
that undertook partial or full automation are likely either to retain or to increase their
overall number of permanent and full-time employees rather than reducing their labor
force. We conducted a chi-squared test, and the results validated the statistical
significance of the association between the two decisions that firms face: χ2(4, N = 1319)
= 18.32, p < .001. Thus, a relationship exists between automation decisions and hiring
practices; in general, firms that adopt automation technologies are more likely to hire
workers.

8
ADBI Working Paper 1081 Gaspar and Harris

Table 3: Hiring and Automation Decisions

No Automated Adopted Partial or
Hiring/Automation Decisions Process Adopted Full Automation Total
Decision 1: Fire 120 (62%) 73 (38%) 193 (100%)
High: 𝐹𝐹𝐹𝐹𝐸𝐸𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶 ≤ – 10% 47 (71%) 19 (29%) 66 (100%)
Low: – 10% < 𝐹𝐹𝐹𝐹𝐸𝐸𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶 < 0 73 (57%) 54 (43%) 127 (100%)
Decision 2: Maintain the status quo 454 (68%) 216 (32%) 670 (100%)
Decision 3: Hire 185 (54%) 157 (46%) 342 (100%)
Low: 0 < 𝐹𝐹𝐹𝐹𝐸𝐸𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶 < 10% 116 (57%) 89 (43%) 205 (100%)
High: 𝐹𝐹𝐹𝐹𝐸𝐸𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶 ≥ 10% 69 (50%) 68 (50%) 137 (100%)
Total 759 (63%) 446 (37%) 1,205 (100%)
Note: Row percentages in parentheses.
Source: Authors’ calculation from World Bank (2015).

We conducted several statistical analyses to assess the baseline model and the
corresponding results. We conducted likelihood ratio tests to check the validity of the
independent variables for inclusion in the baseline model. We found that our choice of
variables to model firms’ hiring decision was statistically significant. We also performed
likelihood ratio tests for combining alternatives to determine whether firms’ hiring choices
are indistinguishable with respect to the variables in the baseline model.
The test showed that we should combine no categories; thus, there are significant
differences between categories. Lastly, our model did not violate the independence of
irrelevant alternatives tests.
Finding 1. Firms that are more likely to automate tend to hire more, ceteris paribus.
The results suggest that firms that have adopted partial or full automation to develop
their new products or services on average hire more workers than their counterpart firms.
Table 4 shows the result of the baseline multinomial logistic regression that models firms’
hiring choices using only three categories, specifically fire, maintain, and hire. The
coefficient estimate indicates that, if a firm decides to automate partially or in full the
processes in its operation that were once conducted manually, the multinomial log-odds
for hiring more employees relative to firing or laying off is likely to increase by 0.3 units
after controlling for other relevant observed time-invariant characteristics. While
statistically non-significant, the coefficient estimate for the status quo choice seems
suggestive of the potential labor market disruption, even in the short run, it being the
least popular choice among firms adopting automation.
The above results show the general pattern that we expected to occur during automation.
We can glean the degree or magnitude of such a pattern from Table 5. Based on the
coefficient estimates, with reference to the choice of more than 10% of worker layoffs,
firms that adopted full or partial automation lean more toward choosing to hire more
employees. It is worth noting that the extent of hiring more than 10% even exhibits the
highest likelihood of being firms’ choice. Automation adoption will indeed cause labor
market disruption in the Philippines, as the results also suggest that maintaining a status
quo, whereby no change in employment will occur, is the least popular option among
firms. This is also generally observable from the results of the alternative model (see
Appendix Table 1), which is robust to firms choosing to hire more than 10%. The results
reveal that the relative probability of hiring rather than laying off in firms with a greater
propensity to automate far outweighs the relative probability of firms that are less likely
to automate, keeping other factors constant.

9
ADBI Working Paper 1081 Gaspar and Harris

Table 4: Result of the Multinomial Logistic Regression of the Firms’ Hiring Model
(Three Categories)
Maintain Status
Quo Hire
Variables Reference = Fire
Automation decision (takes the value 1 if yes and 0 otherwise) –0.227 0.334*
(0.180) (0.194)
Labor productivity (expressed in logs) 0.082 0.101*
(0.052) (0.058)
Firm age 0.018 –0.005
(0.013) (0.013)
Firm age squared 0.000 0.000
(0.000) (0.000)
Firm size (baseline = small)
Medium-sized enterprises –0.419** –0.196
(0.209) (0.230)
Large enterprises –0.218 0.033
(0.232) (0.255)
Industry (baseline = manufacturing)
Retail services 0.608** –0.028
(0.302) (0.345)
Other services 0.247 0.553*
(0.308) (0.320)
Region (baseline = Metro Manila)
Metro Cebu 0.055 –0.085
(0.265) (0.286)
Central Luzon 0.13 –0.208
(0.247) (0.271)
Calabarzon 0.068 –0.439*
(0.218) (0.242)
Constant –0.038 –0.681
(0.744) (0.819)
Log likelihood –1,054.527
LR chi-square 51.361
Number of observations 1090
Standard errors in parentheses.
*** p<0.01, ** p<0.05, * p<0.1.
Source: Authors’ estimates.

10
ADBI Working Paper 1081 Gaspar and Harris

Table 5: Result of the Multinomial Logistic Regression of the Firms’ Hiring Model
(Five Categories)
Maintain
Fire Status Quo Hire
Low No Change Low High
Variables Reference = 𝐅𝐅𝐅𝐅𝐄𝐄𝐂𝐂𝐂𝐂𝐂𝐂𝐂𝐂 ≤– 𝟏𝟏𝟏𝟏%
Automation decision (takes the value of 1 if 0.700** 0.253 0.634** 1.067***
yes and 0 otherwise) (0.344) (0.299) (0.324) (0.34)
Labor productivity (expressed in logs) –0.135 –0.005 0.042 –0.026
(0.096) (0.081) (0.089) (0.094)
Firm age –0.079* –0.054 –0.067 –0.087**
(0.042) (0.04) (0.042) (0.042)
Firm age squared 0.001** 0.001 0.001 0.001**
(0.001) (0.001) (0.001) (0.001)
Firm size (baseline = small)
Medium-sized enterprises 0.482 –0.103 0.411 –0.249
(0.381) (0.318) (0.358) (0.373)
Large enterprises 0.453 0.076 0.658* –0.106
(0.429) (0.358) (0.397) (0.417)
Industry (baseline = manufacturing)
Retail services –0.139 0.516 –0.474 0.264
(0.568) (0.458) (0.540) (0.525)
Other services 0.324 0.462 0.666 0.924
(0.614) (0.541) (0.569) (0.581)
Region (baseline = Metro Manila)
Metro Cebu 0.242 0.231 –0.159 0.391
(0.520) (0.460) (0.501) (0.508)
Central Luzon –0.770* –0.328 –0.755* –0.553
(0.446) (0.359) (0.403) (0.425)
Calabarzon –0.198 –0.065 –0.546 –0.64
(0.403) (0.345) (0.379) (0.414)
Constant 2.900** 2.712** 0.992 1.828
(1.416) (1.206) (1.327) (1.393)
Log likelihood –1,365.4
LR chi-square 92.897
Number of observations 1,090
Standard errors in parentheses.
*** p<0.01, ** p<0.05, * p<0.1.
Source: Authors’ estimates.

To demonstrate clearly the disruptive effect of automation in the domestic labor market,
Table 6 displays the coefficient estimates with the corresponding relative odds,
comparing all possible combinations of choices. Highlighted are the statistically
significant estimates, which suggest that the automation adoption of firms operating in
the Philippines is associated with job creation. The relative probability of hiring is double
that of firing among firms that decided to automate relative to firms that did not, given
that other factors remain constant. It is also the case relative to maintaining the status
quo, suggesting disruptive consequences arising from the automation decision. Note that
firing, albeit low, at less than 10%, has a higher relative probability than maintaining the
status quo among firms that adopted automation.

11
ADBI Working Paper 1081 Gaspar and Harris

Table 6: Relative Odds of Firms’ Hiring Choices

Hiring Choices of Firms Coefficient z P>z Relative Odds
Hire versus fire
5 versus 1 1.067 3.141 0.002 2.908
5 versus 2 0.367 1.394 0.163 1.444
4 versus 1 1.960 0.050 1.886
4 versus 2 –0.066 –0.271 0.787 0.936
Hire versus maintain the status quo
5 versus 3 0.814 4.007 0.000 2.258
4 versus 3 0.382 2.157 0.031 1.464
Maintain the status quo versus fire
3 versus 1 0.253 0.845 0.398 1.288
3 versus 2 –0.447 –2.124 0.034 0.639
Note: We computed the relative odds by exponentiating the estimated coefficients.
Source: Authors’ estimates.

We also computed the difference in the probability of firms’ hiring choices with respect
to their automation adoption. Table 7 shows that the probability of hiring more is on
average about 7 percentage points higher for firms that have adopted automation than
for those that have not. As expected, firms that have automated processes that were
previously completed manually exhibit around a 10 percentage point lower probability of
maintaining the status quo. There are also chances of firing (only at the moderate scale)
during automation, but these results are not statistically significant.

Table 7: Average Marginal Effects (𝝏𝝏𝝏𝝏/𝝏𝝏𝝏𝝏) of Automation on Hiring Choices

Hiring Choices of Firms (𝝏𝝏𝝏𝝏/𝝏𝝏𝝏𝝏) Std Err. z P>z
Decision 1: Fire
High: 𝐹𝐹𝐹𝐹𝐸𝐸𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶 ≤ – 10% –0.025 0.016 –1.58 0.114
Low: – 10% < 𝐹𝐹𝐹𝐹𝐸𝐸𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶 < 0 0.026 0.019 1.37 0.169
Decision 2: Maintain the status quo –0.102 0.031 –3.32 0.001
Decision 3: Hire
Low: 0 < 𝐹𝐹𝐹𝐹𝐸𝐸𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶 < 10% 0.030 0.023 1.31 0.189
High: 𝐹𝐹𝐹𝐹𝐸𝐸𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶 ≥ 10% 0.070 0.020 3.56 0.000
Source: Authors’ estimates.

Finding 2. The likelihood of net job creation associated with the automation decision
differs across industries.
We conducted additional empirical exercises in an attempt to identify the potential
heterogeneity of the above finding. In doing so, we found significantly varying responses
from the sample firms in the “other services” sector, while those of the firms in retail
services exhibit ambiguous responses, given that their coefficient is not statistically
significant. Table 8 shows that only the manufacturing firms display a greater likelihood
of net job creation on adopting partial or full automation in their operation. It could be the
case that the relatively high labor productivity in the manufacturing industry allows firms
to create additional jobs as the output per worker increases. On the other hand, firms
falling into the other services industry, which include those in the hotel and restaurant

12
ADBI Working Paper 1081 Gaspar and Harris

business, logistics, and IT consulting and development, among other related services,
have a higher likelihood of net job losses as they adopt partial or full automation.

Table 8: Sectoral Heterogeneity of Hiring Decisions

Maintain
Fire Status Quo Hire
Low No Change Low High
Variables Reference = 𝐅𝐅𝐅𝐅𝐄𝐄𝐂𝐂𝐂𝐂𝐂𝐂𝐂𝐂 ≤– 𝟏𝟏𝟏𝟏%
Automation decision (takes the value of 1 if 0.897** 0.463 0.848** 1.294***
yes and 0 otherwise) (0.387) (0.341) (0.366) (0.390)
Automation and industry interaction term
Retail services –0.451 –0.199 –0.37 –0.622
(1.165) (0.957) (1.107) (1.076)
Other services –2.145 –2.583** –2.390* –2.223*
(1.365) (1.242) (1.283) (1.302)
Industry (baseline = manufacturing)
Retail services 0.016 0.557 –0.354 0.525
(0.713) (0.555) (0.684) (0.663)
Other services 1.378 1.67 1.825* 2.000*
(1.103) (1.028) (1.055) (1.079)

Labor productivity (expressed in logs) –0.135 –0.005 0.042 –0.026

(0.097) (0.081) (0.089) (0.094)
Firm age –0.081* –0.057 –0.070* –0.089**
(0.042) (0.040) (0.042) (0.042)
Firm age squared 0.001** 0.001 0.001 0.001**
(0.001) (0.001) (0.001) (0.001)
Firm size (baseline = small)
Medium-sized enterprises 0.476 –0.113 0.404 –0.25
(0.381) (0.319) (0.358) (0.374)
Large enterprises 0.464 0.087 0.670* –0.092
(0.429) (0.358) (0.398) (0.417)
Region (baseline = Metro Manila)
Metro Cebu 0.248 0.233 –0.155 0.401
(0.521) (0.461) (0.502) (0.509)
Central Luzon –0.769* –0.328 –0.755* –0.55
(0.446) (0.360) (0.403) (0.426)
Calabarzon –0.217 –0.088 –0.568 –0.659
(0.405) (0.347) (0.381) (0.416)
Constant 2.876** 2.698** 0.971 1.79
(1.414) (1.204) (1.326) (1.390)
Log likelihood –1,362.37
LR chi-square 98.956
Number of observations 1,090
Standard errors in parentheses.
*** p<0.01, ** p<0.05, * p<0.1.
Source: Authors’ estimates.

13
ADBI Working Paper 1081 Gaspar and Harris

Finding 3. Labor market disruptions that deserve policy actions are apparent.
Table 9 summarizes the results after estimating Equation 2. The empirical exercises
reveal that, keeping other factors constant, the expectation is that the job prospects for
non-production workers will remain positive in firms that are more likely to introduce new
production technologies, including those that could automate some tasks. This finding is
intuitive, as non-production workers are highly involved in managerial and administrative
activities as well as sales, fulfilling tasks that are not easily automatable. While there are
already online platforms in the Philippines to conduct selling activities, sales workers
remain relevant for activities that require human contact. Consistent
with the recent findings of Krieger-Boden and Sorgner (2018), we also found an
indicative good prospect for female full-time workers in firms that are more likely to adopt
automation. Largely thanks to the historical expansion in educational attainment, the
female workforce in the Philippines has the potential to handle tasks that may require
higher skills during automation and other technological adoptions at firms across the
state.

Table 9: Propensity to Automate and Employment Structure

Coefficient
Ratio to Total Full-Time Employment Estimates
(Equation 2.1) Non-production workers 0.121***
(0.029)
(Equation 2.2) Female non-production workers 0.000
(0.001)
(Equation 2.3) Unskilled workers 0.080***
(0.030)
(Equation 2.4) Female unskilled workers –0.000
(0.015)
(Equation 2.5) Female full-time workers 0.044***
(0.015)
(Equation 2.6) Temporary or seasonal workers 0.287
(0.186)
(Equation 2.7) Female temporary or seasonal workers 0.089*
(0.043)
Note: Standard errors in parentheses.
*** p<0.01, ** p<0.05, * p<0.1.
Source: Authors’ estimates.

The results also suggest some potential downside risks to the labor market structure.
Table 9 shows that, on average, firms with a higher propensity to automate have a larger
share of low-skilled workers than the total full-time employment. While this may seem
positive, the result also indicates the lower flexibility among lower-skilled workers to
move to other jobs, especially in the instance of full deployment of automated processes.
Bessen et al. (2019) reached similar conclusions following their analysis of the impact of
automation on Dutch workers. They found that automation is less likely to displace lower-
skilled workers, defined as those in the lower wage ranks, than higher-skilled workers.
However, they also pointed out that, if these lower-skilled workers leave work during
automation, they will find it very difficult to source new employment prospects. What is
interesting to note is that, after evaluating the survey data, we found that firms that had
adopted automated processes are more likely to offer formal training programs, which
may provide opportunities for their low-skilled employees to gain and upgrade skills.
Another potential downside from automation based on the results is the associated rise

14
ADBI Working Paper 1081 Gaspar and Harris

in temporary and seasonal workers in firms that are more likely

to automate. These types of workers often lack favorable labor market conditions,
including the necessary support mechanisms and bargaining power. The number of
female workers in this work category is also likely to rise. The societal expectations
of the role of women, especially mothers, in households may explain this trend.
Meanwhile, it is hard to tell whether the rise of the gig economy, which often involves
short-term contracts or freelance work, could explain the expectations of an increased
demand for temporary or seasonal workers. The gig economy in the Philippines, as well
as jobs with flexible work arrangements, like in many other countries, continues to rise
amid the advances in technology and related infrastructure.

6. CONCLUSION AND POLICY IMPLICATIONS

Our study presented a novel way to estimate the potential labor impact of automation,
that is, through the lens of firms. The empirical exercises analyzing firms’ hiring choices
reported that the Philippine labor market outlook remains promising in the age of
automation. The study generally found that the automation revolution can benefit the
labor force, at least in the short run. The productivity effect seems to be at work, offsetting
the possible job losses resulting from automating tasks. Our results reveal that firms that
have adopted automation tend to hire more rather than fewer employees. Indeed, there
is an expectation that firms’ adoption of new and advanced technologies will introduce
disruptions into the labor market. Maintaining the status quo is the least likely outcome
among firms that are considering automating processes.
The automation revolution is also changing the landscape of the country’s labor market.
Non-production workers have good job prospects that could be due to them occupying
roles that are not easily automatable. Further, these workers tend to have higher skills,
making them more flexible and able to carry out other non-automated tasks. Interestingly,
we did not find evidence that female workers will be at a disadvantage once automation
takes place at a wider scope and scale. One concern, however, is the possible trend or
tendency toward temporary and seasonal workers, which may make workers vulnerable
to economic shocks. Why this trend is likely to occur deserves further analysis.
The long-run labor market consequences of further automation remain uncertain. The
short-run overall gains in employment in our findings may not take place in the future
when automation and other more advanced technologies, such as artificial intelligence,
further revolutionize the world of work. The results only reflect short-run labor market
disruptions and cannot explain the longer-term consequences of automation. In addition,
as an earlier section discussed, the potential time mismatch involved in the automation
indicator and firms’ hiring choices does not allow for an interpretation of the results as an
immediate reaction of firms that have automated some or all functions in producing
products or delivering services. Nevertheless, the results give an indication of the net
direction of employment changes.
In the short run, the impact of automation is most likely to vary at the individual level and
thus requires policy intervention. There are workers who are potentially at risk of job
displacement during the process of automation. While income risk will be the most direct
implication of job displacement, there are also adverse consequences for one’s
psychological well-being. Displaced workers are often more at risk of depression,
anxiety, and stress. If support services are available for affected workers, then the
process could be smooth and painless; otherwise, it could have adverse effects on
individuals’ mental health and economic prospects. In this case, the government should

15
ADBI Working Paper 1081 Gaspar and Harris

put in place support services in the form of assistance in the short run and human capital
development in the long run.
An important policy lever in the automation revolution is education and training.
The adoption of new and advanced technologies and processes is inevitable in
profit-maximizing firms. During this process, new tasks may emerge and most likely
will require new skill sets. Workers should possess the right set of skills and
competencies, making them more flexible and therefore more able to occupy new roles.
Automation, studies have revealed, tends to favor cognitive skills, such as social and
communication skills. Thus, the government needs to ensure that the current curriculum
is in line with the set of competencies that firms will demand in the future. Otherwise, the
anxiety over a loss of work in the future would become a reality.

16
ADBI Working Paper 1081 Gaspar and Harris

REFERENCES
Acemoglu, Daron, and Pascual Restrepo. 2018. Artificial Intelligence, Automation and
Work. NBER Working Paper No. 24196. Cambridge, MA: National Bureau of
Economic Research.
Asian Development Bank. 2018. Asian Development Outlook 2018 Theme Chapter—
How Technology Affects Jobs. Mandaluyong City, Philippines: Asian
Development Bank.
Autor, David H. 2015. “Why Are There Still So Many Jobs? The History and Future of
Workplace Automation.” Journal of Economic Perspectives 29 (3): 3–30.
Bessen, James. 2015. Learning by Doing: The Real Connection between Innovation,
Wages, and Wealth. New Haven and London: Yale University Press.
Bessen, James, Maarten Goos, Anna Salomons, and Wiljan van den Berge. 2019.
Automatic Reaction—What Happens to Workers at Firms that Automate?
Law & Economics Series Paper No. 19-2. Boston, MA: Boston University
School of Law.
Chang, Jae-Hee, Gary Rynhart, and Phu Huynh. 2016. ASEAN in Transformation: How
Technology is Changing Jobs and Enterprises. Bureau for Employers’ Activities
Working Paper No. 10. Geneva: International Labour Office.
Ford, Martin. 2015. The Rise of the Robots: Technology and the Threat of a Jobless
Future. New York: Basic Books.
Innes, Mike, and Ben Morrison. 2017. “Projecting the Future Impact of Advanced
Technologies: Will a Robot Take my Job?” InPsych: The Bulletin of the
Australian Psychological Society Ltd 39 (2): 34–35.
International Federation of Robotics. 2018. “Robot density rises globally.”
https://ifr.org/ifr-press-releases/news/robot-density-rises-globally.
Krieger-Boden, Christiane, and Alina Sorgner. 2018. “Labor Market Opportunities for
Women in the Digital Age.” Economics: The Open-Access, Open- Assessment
E-Journal 12 (28): 1–8. http://dx.doi.org/10.5018/economics-ejournal.ja.2018-
28.
McKinsey Global Institute. 2017. A Future that Works: Automation, Employment and
Productivity. San Francisco: McKinsey Global Institute.
Orozco, Linda V. 2017. ASEAN in Transformation: How Technology is Changing
Jobs and Enterprises. The Philippines Country Brief. Geneva: International
Labour Office.
Schwab, Klaus. 2016. The Fourth Industrial Revolution. Geneva: World Economic
Forum.
World Bank. 2016. World Development Report 2016: Digital Dividends. Washington,
DC: World Bank.

Dataset

The World Bank. 2015. Philippine Enterprise Survey. Data available at

http://www.enterprisesurveys.org.

17
ADBI Working Paper 1081 Gaspar and Harris

APPENDIX
Appendix Table 1: Robustness Check: Multinomial Logistic Regression
of the Firm Hiring Model Using the Index of Automation Propensity
Maintain
Fire Status Quo Hire
Low No Change Low High
Variables Reference = 𝐅𝐅𝐅𝐅𝐄𝐄𝐂𝐂𝐂𝐂𝐂𝐂𝐂𝐂 ≤– 𝟏𝟏𝟏𝟏%
Index of automation propensity (1 being the 0.063 -0.37 0.553 2.100***
highest and 0 the lowest) (0.741) (0.637) (0.686) (0.716)
Labor productivity (expressed in logs) -0.105 0.019 0.059 -0.043
(0.097) (0.082) (0.090) (0.096)
Firm age -0.072* -0.049 -0.061 -0.078*
(0.042) (0.040) (0.042) (0.041)
Firm age squared 0.001* 0.001 0.001 0.001*
(0.001) (0.001) (0.001) (0.001)
Firm size (baseline = small)
Medium-sized enterprises 0.434 -0.115 0.368 -0.406
(0.380) (0.319) (0.358) (0.374)
Large enterprises 0.453 0.115 0.631 -0.334
(0.438) (0.367) (0.406) (0.427)
Industry (baseline = manufacturing)
Retail services -0.139 0.512 -0.467 0.322
(0.567) (0.457) (0.539) (0.525)
Other services 0.288 0.429 0.65 0.998*
(0.613) (0.541) (0.568) (0.579)
Region (baseline = Metro Manila)
Metro Cebu 0.157 0.235 -0.192 0.354
(0.518) (0.459) (0.500) (0.507)
Central Luzon -0.813* -0.366 -0.756* -0.466
(0.445) (0.359) (0.403) (0.423)
Calabarzon -0.247 -0.091 -0.528 -0.594
(0.403) (0.346) (0.380) (0.415)
Constant 2.699* 2.495** 0.787 1.853
(1.412) (1.209) (1.328) (1.395)
Log likelihood -1,372.3
LR chi-square 106.5
Number of observations 1,103
1
Using the estimated propensity to automate index as an alternative indicator for automation.
Standard errors in parentheses.
*** p<0.01, ** p<0.05, * p<0.1.
Source: Authors’ estimates.

18