Using Machine Learning to Measure Conservatism

Download or compute conservatism score

Download Conservatism Score

Data is available in comma-separated (CSV), or parquet formats. These files contain all the measures used in the paper as described in the table below.
Measure Estimate Features set (covariates)
L1 Linear Features set 1
L2 Linear Features set 2
L1y Linear by fiscal year Features set 1
L2y Linear by fiscal year Features set 2
MLC1_mean Average of 100 Neural Network models Features set 1
MLC2_mean Average of 100 Neural Network models Features set 2
MLC1y_mean Average of 100 Neural Network models Features set 1 and years fixed effects
MLC2y_mean Average of 100 Neural Network models Features set 2 and years fixed effects
MLC1 Neural Network trained on MLC1_mean Features set 1
MLC2 Neural Network trained on MLC2_mean Features set 2
MLC1y Neural Network trained on MLC1y_mean Features set 1 and years fixed effects
MLC2y Neural Network trained on MLC2y_mean Features set 2 and years fixed effects

Features set 1 includes Size, Market-to-Book ratio (M/B), Leverage and features set 2 adds Non-Operating Accruals, Cashflow from Operating Activities, Investment Cycle, Firm Age.






Compute the C score

You can use the following utility to compute the C score in your browser.

Note: This utility uses your own compuer ressources to compute the C score and is limited in the size of the file it accepts. We suggest that you only provide a file with the columns necessary to compute the C score.

Select a CSV file containing the variables detailed in the table below (formulas use COMPUSTAT and CRSP variables names):

{{message_data}}

Select columns for


Variable Description Formula / Note
Identifiers Typically gvkey/permno and fiscal year. Use 'Ctrl' to select multiple identifiers.
Size Natural logarithm of the market value of equity corrected for inflation. $\log(\frac{prcc\_f \times csho}{gdpdef})$
M/B ratio Ratio of market value of equity to book value of equity. $\frac{prcc\_f \times csho}{ceq}$
Financial Leverage Long-term debt plus short term debt deflated by market value of equity. $\frac{dltt + dlc}{prcc\_f \times csho}$
Non-Operating Accruals Non-Operating Accruals scaled by lagged assets. $\frac{(ib+dp)-cfo-oacc}{at_{-1}}$
CFOA Cash Flow from operating activities. $\frac{oancf}{at_{-1}}$ if $fyear \ge 1987$, $\frac{fopt-oacc}{at_{-1}}$ if $fyear<1987$.
Investment Cycle Depreciation expenses deflated by lagged assets. Standard deviation of $\frac{capx}{at}$ over the past 5 years (3 years minimum).
Age Age of the firm in years. Number of years with return history on CRSP or number of years in Compustat if not available.
Volatility Standard deviation of daily returns over the past year. $std(ret_d)$ where $ret_d$ is the series of daily returns over the past 253 trading days.


NOTE: Processing can take some time for large files. Please do not lose faith ...


{{ message_cscore }}

{{ message_download}}

Additional Information

The code used to generate the C Score and the paper's results can be viewed here.