Ever thought about how many people in the Eastern Suburbs of Sydney NSW are employed as Chefs? Or how many people living in Cairns in Queensland are employed as Sales Assistants?
These are the questions the National Skills Commission’s (NSC) new experimental Nowcast of Employment by Region and Occupation (NERO) is seeking to help answer. Our Nowcasting and Economic Modelling team has produced NERO using nowcasting, an emerging method for evaluating present conditions, leveraging big data and machine learning techniques. NERO provides monthly estimates of employment for 355 occupations (at the ANZSCO 4-digit level) across 88 regions (at the SA4 level), or over 31,000 series in total.
Explore the data dashboard here Nowcast of Employment by Region and Occupation
Before the release of NERO, the five-yearly ABS Census of Population and Housing was Australia’s key dataset in terms of detailed labour market information. The Census, as well as the ABS Labour Force Survey, are both invaluable in providing snapshots of the labour market at a point in time, as well as how it is changing over the medium term. But it is difficult to gauge detailed occupational and regional employment trends (such as at the ANZSCO-4 digit and SA4 level) on a more frequent basis using these sources. This gap in labour market estimates is what NERO can help address.
How were the NERO nowcasting estimates derived?
Nowcasting is an emerging methodology that is used to bridge the gap between economic indicators by ‘predicting the present’. Nowcasting does this by blending big data and more traditional datasets with machine learning and other modelling techniques to create frequent, timely and granular estimates. While most nowcasting literature examines nowcasting approaches for aggregate or economy-wide indicators (for example, total GDP or total unemployment rates), NERO extends upon this by providing new experimental nowcasting estimates at detailed occupation and regional levels.
NERO draws on a wide range of datasets to form inputs into the model. This includes the ABS Census of Population and Housing, Labour Force Survey and National Accounts, online job advertisements, data from the NSC Internet Vacancy Index and Burning Glass Technologies, job placement data from the Australian Government’s ‘jobactive’ program, data on visa holders from the Department of Home Affairs, as well as other customised data sources from the ABS. Once these datasets were collected and processed for use, including checking reference release dates, reference periods and ANZSCO/SA4 concordances, three well-known machine learning modelling techniques were then applied to develop estimates of employment – namely, Random Forest, Gradient Boosting and Elastic Net Regression models. The results of these applications were then combined (or stacked) to produce a single optimal set of nowcasts of employment. Performance of the model was evaluated against the results of previous Censuses, as well as customised data from the ABS. The nowcasts were also smoothed and scaled to ensure they are broadly consistent with the more aggregated labour market data published by the ABS Labour Force Survey.
Further information on the methodology can be found here: Nowcasting methodology
What are the main benefits of the new NERO dataset?
NERO provides a new type of evidence in three fundamental ways:
1. Granularity of employment estimates
By granularity, we mean we can target specific occupations (4-digit level ANZSCO codes) and regions (based on SA4 geographies). This is best illustrated by way of example. Following the initial enactment of nationwide restrictions due to the onset of COVID-19 in March 2020, using NERO we find two different occupational employment trends within the same industry in Sydney – Eastern Suburbs (NSW). Looking at the NERO data in Figure 1, we see that the number of people living in the Sydney – Eastern Suburbs who are employed as Bar Attendants and Baristas begins trending upwards from May 2020. By contrast, the number of people in the region who are employed as Chefs slowed in June, before slipping further over the following months. Without the level of granularity provided by NERO, we would not be able to suggest differences in employment recovery rates between occupations in hospitality within the Eastern Suburbs of Sydney.
|Date||Bar Attendants and Baristas||Chefs|
2. Frequency of employment estimates
As the Census is conducted on a five-year basis, it is difficult to use Census data to identify potential economic turning points that occur within these five-year gaps. NERO helps to bridge this gap by providing monthly estimates. This enables turning points to be more easily identified, particularly over time periods of between six and twelve months where changes in trends can be more confidently identified. For example, through NERO we can describe the recovery trends of employment for Bar Attendants and Baristas in Sydney – Eastern Suburbs (NSW) as starting in May of 2020 (indicated by the diamond in Figure 2). With the frequency of monthly estimates NERO can support the analysis of new and emerging developments in the labour market, particularly during times of labour market change.
|Date||NERO Final Estimated Employment|
3. Timeliness of employment estimates
The release of NERO is also an important step towards providing a timelier view of employment. Timely data is data that has a short delay from its observed reference period and its eventual release, supporting analysts and policy makers to develop policies that consider current labour market conditions.
There is always a trade-off between having extremely timely estimates that are prone to incorrect (or false) signals, versus having more reliable (but less timely) trend estimates. To balance the trade-off between accuracy and timeliness, the NERO data have been smoothed to provide an indication of trends in local labour markets. Smoothing of NERO is required as the raw predictions from the underlying model at the more finely detailed SA4 and ANZSCO 4-digit levels, can display significant variability (or incorrect signals), given some series have very small numbers of people employed. A limitation of using such a smoothing filter, however, is that the current experimental NERO estimates may not perfectly reflect very recent changes in the labour market, such as the impacts of lockdown measures currently being experienced by many regions across the country.
It is in times like the present lockdown that analysts and policy makers are advised to combine NERO estimates with all other available evidence, as they seek to understand what is happening in local labour markets.
What can NERO tell us about the impacts of recent lockdowns?
While we would not recommend interpreting raw predictions from NERO at this stage (given the inherent volatility in some series), we note that some of the raw predictions can in fact pick up shorter-term employment shocks such as those arising in lockdowns. This is illustrated in the latest release for Chefs in Sydney – Eastern Suburbs. In this case, it is likely that the dramatic drop in raw predictions from June 2021 to July 2021 corresponds with the start of the current lockdowns in Sydney, like that experienced in early-mid 2020.
|Date||NERO Final Estimated Employment||Raw Predictions|
The NSC is hopeful incorporating new data sources into NERO, validating the NERO estimates against new information (such as the forthcoming data from the 2021 Census) and improving the underlying smoothing methodologies may enable more timely and responsive NERO estimates to be published in the future.
As a new dataset, NERO adds to the sources of information available to help analyse and understand Australia’s labour market. NERO complements existing sources and provides new experimental estimates that seek to balance the need for accuracy, granularity, frequency, and timeliness. NERO also represents methodological advances and shows how cutting-edge data analytics, involving the use of big data and machine learning techniques in nowcasting, can be applied in a labour market context. As NERO is still experimental in nature, there are still areas of improvement for the model – including incorporating new data sources, taking on board feedback from stakeholders and validating the estimates against new information as it becomes available (such as the forthcoming data from the 2021 Census).