RDP 2023-03: Doing Less, with Less: Capital Misallocation, Investment and the Productivity Slowdown in Australia 3. Data

The analysis in this paper exploits firm-level data from the Business Longitudinal Analysis Data Environment (BLADE), compiled by the Australian Bureau of Statistics. BLADE consists of administrative data from the Australian Taxation Office (ATO) on the near universe of firms matched with ABS survey microdata, such as the Business Characteristics Survey.

The particular data we use come from firms' business income tax forms and pay as you go (PAYG) employment forms. The former contain data on firms' sales, income and expenses, and balance sheet. These are used to construct measures of firm output, inputs and capital stocks. The PAYG statements are used to derive a measure of full-time equivalent (FTE) employment (Hansell, Nguyen and Soriano 2015), which we use as labour inputs.

For most of the analysis, our measure of capital is defined as (deflated) non-current assets. As discussed below, we also construct a measure of the capital stock using a perpetual inventory method (PIM), which uses information on the firm's investment, depreciation and asset sales to build an estimate of the capital stock (see Appendix A).

Our main measure of productivity is MFP, as this is the more theoretically appropriate metric (Decker et al 2020). MFP is obtained by estimating a translog, gross output production function, using the proxy approach in Ackerberg, Caves and Frazer (2015). For more details see Hambur (forthcoming). We also construct simple measures of labour productivity (ratio of value-added to FTE) and capital productivity (ratio of value-added to capital stock), for extensions and robustness.

All production function inputs are deflated using division-level deflators rather than firm-level prices. While not ideal, this is very common in the literature as firm-level prices are not readily available in most datasets. Moreover, several papers have demonstrated that measures of productivity deflated using industry- and firm-level deflators tend to be very similar, and lead to similar results in regressions such as those undertaken in this paper (Foster, Haltiwanger and Syverson 2008; Decker et al 2020; Blackwood et al 2021).^[4]

While BLADE contains the (near) universe of Australian firms, we have to make some exclusions. We focus on the non-financial market sector, as is common in the literature, given the difficulty measuring capital and outputs in the finance sector and complexities around the role of government in the public sector. We focus on employing firms, given the need for non-zero labour inputs in productivity estimation. And we remove sole trader organisations, given these organisations do not report on their balance sheets in the tax data, as well as very young firms given we need a few years of data to estimate MFP (though these are included in some extensions using capital productivity). We include the primary sector (agriculture and mining) in our analysis, but show that the results are robust to its exclusion.

Even with these exclusions we capture a very large proportion of the non-finance market sector: on average we cover about 60 per cent of the sales in each constituent industry division (see Hambur (forthcoming)). As such the dataset remains large and highly representative, though it will be slightly skewed towards older and larger firms compared to Andrews and Hansell (2021).

Footnote

Our measure is equivalent to TFPR^rr in Blackwood et al (2021). [4]