The Jones (1991) Model Setting Revisited

What would Jennifer J. Jones do, today?

Disclaimer

I am not Jennifer J. Jones and (unfortunately) have never talked to her either. The following analysis is my own interpretation and adaptation of her seminal 1991 paper on earnings management using discretionary accruals. Any errors or misinterpretations are solely my responsibility. The sole purpose of this exercise is to illustrate how one might revisit and extend classic empirical accounting research using modern econometric techniques and data availability. Also, it is meant to be a fun exercise in applied econometrics. Last but not least, this is my way of learning and generating ideas for my own research. Please do not cite or circulate this work without my permission.

Empirical Strategy

Sample Selection

Matching Import Relief Investigations to Industries and Compustat Firms

I identify import relief investigations from the U.S. International Trade Commission (USITC) by scraping their Commission Publications Library. My scraping process employs robust retry logic and Cloudflare bypass mechanisms, with raw HTML cached for reproducibility and efficiency. I match investigation titles to standardized industry classifications using a SIC-NAICS crosswalk (1987/1997 versions) enhanced by semantic matching via a local language model. This procedure links (almost) each USITC investigation ID to corresponding SIC and NAICS codes, enabling systematic identification of treatment industries (those under investigation) and control industries (structurally similar but not investigated). For implementation details, see the companion Python scripts. The table below summarizes the matched import relief investigations and their corresponding industries.

Matched Import Relief Investigations and Corresponding Industries
Investigation Title	SIC Title	SIC Code	LLM Reasoning
Crystalline Silicon Photovoltaic Cells	Nonclassifiable Establishments	9999	None of the listed industries correspond to photovoltaic cell manufacturing; thus the case is best classified as nonclassifiable.
Large Residential Washers	Bolts, Nuts, Screws, Rivets, and Washers	3452	The keyword "washers" matches the SIC 3452 industry for bolt, nut, screw, rivet, and washer manufacturing, and its NAICS counterpart 332722.
Leather Wearing Apparel	Leather Goods, NEC	3199	The title refers to apparel made of leather, aligning best with leather goods manufacturing (SIC 3199.0 / NAICS 316999.0).
Fish	Fish and Seafoods	5146	The title directly references fish, matching the Fish and Seafoods wholesale industry.
Fresh Cut Roses	Fresh Fruits and Vegetables	5148	Fresh cut roses are a type of fresh produce best matched with the Fresh Fruits and Vegetables Wholesalers industry.
Motor Vehicles and Chassis and Bodies Thereof	Truck and Bus Bodies	3713	The case title focuses on motor vehicles, chassis, and bodies, matching the Motor Vehicle Body Manufacturing industry.
Fishing Rods and Parts Thereof	Metal Shower Rods	3432	Fishing rods are a type of metal (or fabricated) rod product, best fitting the Metal Shower Rods manufacturing industry and its associated NAICS category of miscellaneous fabricated metal products.
Tubeless Tire Valves	Industrial Valves	3491	Tubeless tire valves are a type of industrial valve component, best fitting the Industrial Valve Manufacturing category.
Heavyweight Motorcycles	Motorcycles, Bicycles, and Parts	3751	Direct match to motorcycle manufacturing.
Stainless Steel and Alloy Tool Steel	Steel Foundries, NEC	3325	Stainless and alloy tool steels are typically produced in steel foundries, matching the SIC 3325 and NAICS 331513 categories.
Nonrubber Footwear	Footwear, Except Rubber, NEC	3149	Direct match to nonrubber footwear manufacturing.
Carbon and Certain Alloy Tool Steel Products	Steel Foundries, NEC	3325	The case involves manufacturing of specialized tool steel, which is typically produced in steel foundries, matching SIC 3325 and NAICS 331513.
Unwrought Copper	Primary Smelting and Refining of Copper	3331	Unwrought copper refers to the raw refined copper produced in primary smelting and refining operations.
Canned Tuna Fish	Canned and Cured Fish and Seafood	2091	Direct match to canned and cured fish and seafood industry.
Potassium Permanganate	Nonclassifiable Establishments	9999	Potassium Permanganate is a chemical product not covered by the listed manufacturing industries, so it falls under the Nonclassifiable/Unclassified category.
Nonrubber Footwear	Footwear, Except Rubber, NEC	3149	Directly matches the industry for footwear manufacturing excluding rubber products.
Wood Shingles and Shakes	Shingle Mills, Shakes	2429	The case title refers to shingle manufacturing, directly matching the Shingle Mills (Shakes) industry and its NAICS classification as sawmills.
Electric Shavers and Parts Thereof	Electronic Parts and Equipment, NEC	5065	Electric shavers are small electronic appliances, fitting best with the generic electronic parts and equipment industry.
Metal Castings	Nonferrous Die-Castings, Except Aluminum	3364	The title refers to metal casting, which aligns best with nonferrous die-casting foundries (excluding aluminum).
Apple Juice	Apple Orchards and Farms	175	Apple Juice is produced from apples, making the apple orchard industry the most directly relevant among the provided options.
Steel Fork Arms	Steel Foundries, NEC	3325	Steel fork arms are steel components manufactured by a foundry, matching the Steel Foundries, NEC industry.
Knives	Lawnmower Repair Shops, Sharpening and Repairing Knives, Saws and Tools	7699	The industry description for SIC 7699.0 lists 'Sharpening and Repairing Knives, Saws and Tools', directly matching the case title.
Certain Cameras	Nonclassifiable Establishments	9999	No candidate industry matches the keyword 'cameras'; therefore the nonclassifiable category is chosen.
Extruded Rubber Thread	Molded, Extruded, and Lathe-Cut Mechanical Rubber Goods	3061	The product is an extruded rubber thread, matching the description of Molded, Extruded, and Lathe-Cut Mechanical Rubber Goods and its NAICS counterpart for rubber products manufactured for mechanical use.
Fresh Winter Tomatoes	Fresh Fruits and Vegetables	5148	The title refers to fresh tomatoes, directly matching the Fresh Fruits and Vegetables industry and its wholesale distribution.
Broom Corn Brooms	Brooms and Brushes	3991	The title refers to manufacturing brooms, matching the Brooms and Brushes industry.
Fresh Tomatoes and Bell Peppers	Fresh Fruits and Vegetables	5148	The case title references fresh produce (tomatoes and bell peppers), matching the Fresh Fruits and Vegetables wholesaler industry.
Wheat Gluten	Wheat	111	Wheat gluten is a product derived directly from wheat, so the most appropriate industry is wheat farming.
Lamb Meat	Meat Markets	5421	The title refers to retail sale of lamb meat, matching Meat Markets.
Steel Wire Rod	Steel Wire Drawing	3315	Direct match to steel wire drawing, which produces wire rods.
Certain Circular Welded Carbon Quality Line Pipe	Steel Pipe and Tubes	3317	The product is a welded carbon-quality line pipe, which is a steel pipe manufactured from purchased steel.
Crabmeat from Swimming Crabs	Nonclassifiable Establishments	9999	No provided industry relates to seafood or crab processing, so the case falls into the nonclassifiable/unclassified category.
Extruded Rubber Thread	Molded, Extruded, and Lathe-Cut Mechanical Rubber Goods	3061	The title indicates extruded rubber threads, which are mechanical rubber goods produced by extrusion, matching SIC 3061 and NAICS 326291.
Steel: Determinations and Views of Commissioners	Steel Foundries, NEC	3325	The case concerns the steel industry in general, best aligning with steel foundry manufacturing.
Crystalline Silicon Photovoltaic Cells	Nonclassifiable Establishments	9999	No industry in the list matches photovoltaic cell manufacturing; select the nonclassifiable/unclassified category.
Large Residential Washers	Bolts, Nuts, Screws, Rivets, and Washers	3452	The only industry listed that explicitly mentions washers is Bolt, Nut, Screw, Rivet, and Washer Manufacturing, making it the closest match.
Fresh	Fresh Fruits and Vegetables	5148	The title 'Fresh' directly matches the description of Fresh Fruits and Vegetables and Fresh Fruit and Vegetable Wholesalers.
Fine Denier Polyester Staple Fiber	Broadwoven Fabric Mills, Manmade Fiber and Silk	2221	Polyester staple fiber is a manmade fiber used in broadwoven fabric manufacturing, matching the Broadwoven Fabric Mills industry.

Some important caveats about the matching process:

I only use safeguard investigations (not other types of import relief)
Some investigations could not be matched to a specific SIC code (9999)
The matching relies on LLM reasoning which may introduce errors

Accounting Data

I obtain financial statement data from Compustat North America for U.S. incorporated firms beginning in 1975. I exclude financial institutions (SIC codes 6000-6999) and require non-missing total assets for all firm-years. To ensure robust industry-specific estimation of the Jones model, I retain only industry-years (three-digit SIC level) with at least 10 observations. I exclude observations with missing values for total accruals, revenue changes, property-plant-equipment, or lagged total assets. All continuous variables are winsorized at the 1st and 99th percentiles to reduce the influence of extreme outliers. For additional implementation details, see the companion R script.

Treatment Sample Selection and Event Window Definition

To construct my treatment sample, I apply several critical filters to the matched investigations. First, I identify the investigation date and extract both the calendar year and month. To ensure that year_zero captures the last complete fiscal year before the investigation announcement, I adjust for each firm’s fiscal year-end month. Specifically, for firms with fiscal year-end months before the investigation month, their fiscal year statement would have already been reported before the investigation, making that fiscal year “year 0.” For firms with fiscal year-end months after the investigation month, their reported year 0 encompasses the investigation announcement, and I adjust their timing accordingly. Second, I address the issue of multiple investigations in the same industry. Some SIC-3 industries (314, 332, 514) experience multiple import relief investigations within short time windows. To avoid contamination and confounded treatment effects, I retain only the first investigation per SIC-3 industry, excluding subsequent treatments within the same year or immediately following year (diff_year_zero > 1). This design choice prioritizes clean identification of the initial shock at the cost of sample size—a conservative approach that trades statistical power for causal credibility.

The OG… Jones Model

Jones (1991): Time-Series Firm-Specific Approach

\[ \frac{\text{Total accruals}_{it}}{AT_{i,t-1}} = \frac{1}{AT_{i,t-1}} + \beta_1 \frac{\Delta REV_{it}}{AT_{i,t-1}} + \beta_2 \frac{PPE_{it}}{AT_{i,t-1}} + \epsilon_{it}\]

Where:

\(\frac{\text{Total accruals}_{it}}{AT_{i,t-1}}\) = total accruals for firm i at time t
\(AT_{i,t-1}\) = total assets for firm i at time t-1
\(\Delta REV_{it}\) = change in revenues for firm i at time t
\(PPE_{it}\) = property, plant, and equipment for firm i at time t

Jones (1991) estimates a separate regression for each firm using several years of historical data. The estimation period stops in year t-2. Firm-specific coefficients capture each firm’s unique accrual patterns. Compares actual accruals to the firm’s own historical norm.

Time-Series Firm-Specific Jones Model

Discretionary accruals: \(\text{DA}_{it}\) = residual from firm i’s time-series regression

My Approach: Cross-Sectional Industry-Specific Approach

Estimates yearly regressions by SIC-3 industry using all firms in that industry. Industry-year-specific coefficients capture industry norms in each period. Compares firm accruals to industry peers in the same year.

Cross-Sectional Industry-Specific Jones Model

\(\text{DA}_{it}\) = residual from industry-year cross-sectional regression

Implication: I measure abnormal accruals relative to contemporaneous industry peers, not firm’s historical pattern

Deep Dive: Earnings Management Measurement

Dechow et al. (2010) reviews the literature surrounding “earnings quality,” including earnings management. They discuss trade-offs between time-series and cross-sectional approaches to measuring abnormal accruals. See Gow and Ding (2024) for an in-depth overview of implementing accrual-based earnings management measures—or for practical guidance on many aspects of empirical research in accounting.

Stacked Difference-in-Differences with First-Differenced Specification

I implement a stacked difference-in-differences (DiD) design to estimate the impact of import relief investigations on discretionary accruals. The treatment group consists of firms in industries subject to import relief investigations, while the control group includes firms in industries without such investigations (different SIC-1 codes). I define event time relative to the investigation year, with t=0 as the investigation year. For each event window (t-2 to t+2), I create a separate dataset stacking observations from treated and control industries; see, e.g., Baker et al. (2022); Baker et al. (2025) for more details on stacked DiD designs.

First-Difference DiD Model

\[\Delta \text{DA}_{it} = \beta_1Treated_i + \alpha_{fyear}+ \gamma_{cohort}+ \epsilon_{it}\]

Where:

\(\Delta(DA_{it}) = \text{DA}_{it} - \text{DA}_{i,t-1}\) is the change in discretionary accruals for firm i at time t
\(Treated_i\) is a binary indicator for whether firm i is in the treatment group
\(\alpha_{fyear}\) are fixed effects for the fiscal year
\(\gamma_{cohort}\) are fixed effects for the cohort
\(\epsilon_{it}\) is the error term

Why?

The idea is to present results in a way that is more directly comparable to Jones (1991). She also compares changes in discretionary accruals around the investigation period, specifically in years t-1 to t+1. I want to make my results more directly comparable to hers.

Okay, so where’s the DiD?

Difference #1: Change in discretionary accruals within firms over time. First differencing removes firm fixed effects.

Difference #2: Difference in changes between treated and control industries over time. The coefficient on \(Treated_i\) captures the average difference in changes in discretionary accruals between treated and control industries.

Results

Summary Statistics

Summary Statistics for Key Variables
Variable	N	Mean	SD	Min	25th Pct	Median	75th Pct	Max
Total Accruals	157866	-0.05	0.22	-2.53	-0.09	-0.04	0.00	1.19
Discretionary Accruals	157866	0.01	0.19	-2.61	-0.04	0.00	0.05	2.77
Total Assets	157866	1843.36	9560.28	0.00	19.27	114.32	815.61	551669.00
Inverse Lagged Total Assets	157866	0.14	0.61	0.00	0.00	0.01	0.06	8.00
Change in Revenues	157866	0.14	0.41	-1.17	0.00	0.07	0.23	2.99
Property, Plant, and Equipment	157866	0.78	0.52	0.00	0.37	0.71	1.15	3.14
Treated Indicator	157866	0.01	0.11	0.00	0.00	0.00	0.00	1.00
Time Relative to Investigation	157866	-0.03	1.99	-3.00	-2.00	0.00	2.00	3.00
Fiscal Year	157866	1989.79	11.42	1976.00	1981.00	1986.00	1997.00	2023.00

First-Difference DiD Estimates

The table and figure below present the results of the first-difference DiD estimation. The table shows the estimated treatment effects for each time window around the import relief investigations, while the figure plots the coefficients for each estimation window to visualize the dynamics of changes in discretionary accruals between treated and control industries for each time window.

First-Difference DiD Estimates of Changes in Jones Residuals Around Import Relief Investigations

Event-study: Difference-in-Differences Estimates of Jones Residuals Around Import Relief Investigations

Limitations

What about other (real) earnings management measures?

Typically, one would like to see results using multiple measures of earnings management; see, e.g., Dechow et al. (2010) for an overview - Typically, one would like to see results using multiple measures of earnings management, see e.g. Dechow et al. (2010) fot an overview - Breuer and Schütt (2023) provides a Bayesian framework to combine multiple earnings management measures into a single latent construct - To go even further, one could consider real earnings management measures as in Roychowdhury (2006) (better leave that up to Reviewer 2 to decide for you…) - While you are at it, also check and apply Srivastava (2019) for improved measurement of real earnings management (I see you, Reviewer 2…)

Residuals as Dependent Variable

Discretionary accruals are estimated residuals from the Jones model (two-step approach)
Chen et al. (2018) and B. Chen et al. (2023) highlight potential bias and inefficiency in two-step approaches
- Solution: Use control function or joint estimation methods
I am not giving this a separate headline, but make sure your standard errors game is on par with the latest literature (e.g., W. Chen et al. (2023) or Abadie et al. (2023)), in case Reviewer 2 asks you about it…

Statistical Power

Effect is statistically significant (p < 0.05) in one period—where we would expect it (\(t=0\)) to be strongest
Not statistically significant in other periods
Standard errors are clustered at the cohort-gvkey level—may be conservative but necessary to account for within-firm and within-cohort correlation (I see you, Reviewer 2…)

Economic Magnitude

Effect size (2.0 pp) is approximately 1/3 of Jones’s finding
Could reflect: (1) fiscal year dilution, (2) industry aggregation, (3) control group contamination, (4) measurement error in cross-sectional DA models

Control Group Validity

Control industries (different SIC-1) may still face import competition
Could attenuate treatment effect if controls also manipulate
Parallel trends are validated in the pre-period, but the assumption may not hold throughout

Conclusion

The findings provide evidence consistent with Jones (1991) using a complementary identification strategy. The stacked DiD with cross-sectional discretionary accruals and contemporaneous industry controls offers:

Stronger causal identification through parallel trends validation
Robustness to systematic time-period effects
Industry-level confirmation of firm-level manipulation documented by Jones

References

Abadie, A., Athey, S., Imbens, G.W., Wooldridge, J.M., 2023. When should you adjust standard errors for clustering? The Quarterly Journal of Economics 138, 1–35.

Baker, A., Callaway, B., Cunningham, S., Goodman-Bacon, A., Sant’Anna, P.H., 2025. Difference-in-differences designs: A practitioner’s guide. arXiv preprint arXiv:2503.13323.

Baker, A.C., Larcker, D.F., Wang, C.C., 2022. How much should we trust staggered difference-in-differences estimates? Journal of Financial Economics 144, 370–395.

Breuer, M., Schütt, H.H., 2023. Accounting for uncertainty: An application of bayesian methods to accruals models. Review of Accounting Studies 28, 726–768.

Chen, B., Chen, W., Hribar, P., Melessa, S., 2023. Measuring dual effects in accounting research: Limitations of absolute residuals and the role of quantile regression. https://doi.org/10.2139/ssrn.4572590

Chen, W., Hribar, P., Melessa, S., 2023. Standard error biases when using generated regressors in accounting research. Journal of accounting research 61, 531–569.

Chen, W., Hribar, P., Melessa, S., 2018. Incorrect inferences when using residuals as dependent variables. Journal of Accounting Research 56, 751–796.

Dechow, P., Ge, W., Schrand, C., 2010. Understanding earnings quality: A review of the proxies, their determinants and their consequences. Journal of accounting and economics 50, 344–401.

Gow, I.D., Ding, T., 2024. Empirical research in accounting: Tools and methods. Chapman; Hall/CRC.

Jones, J.J., 1991. Earnings management during import relief investigations. Journal of accounting research 29, 193–228.

Roychowdhury, S., 2006. Earnings management through real activities manipulation. Journal of accounting and economics 42, 335–370.

Srivastava, A., 2019. Improving the measures of real earnings management. Review of Accounting Studies 24, 1277–1316.