Hyperinflation gives an interesting twist and I’m only familiar with how to handle it in financial statements at a high level. Of course a major problem with recent statements is that assets may be carried at historical values, and if you translate them into dollars at the current rate they look like they’re worth far less than they really are. To bring balance sheets up to date you have to go through the pain of multiplying items that are carried at historical cost (usually non-monetary) by the change in price index since their time of purchase.

From a pure forecasting EPS point of view, my best guess would be that if you had the historical inflation data, you could subtract it out from EPS growth in order to get the real growth. Then you’d need to add back in your best guess for inflation for the next year or two in order to forecast future EPS. Have you tried something like this?

Best,
Lumi

Here is an example using the current numbers for what looks to be a highly predictable business, PAYX. Right now, the last 10 years EPS’s were:

.51 .68 .73 .78 .8 .97 1.22 1.35 1.56 1.48

If you compute what the % gain in EPS is each year, you get:

33.3% 7.4% 6.8% 2.6% 21.3% 25.8% 10.7% 15.6% -5.1%

Now the mean of the above % gains is 13.1% and the standard deviation is 12.1%.

If the returns are normally distributed, then 68% of the returns should be between +/- 1 standard deviation from the mean and 95% of the returns should be between +/- 2 standard deviations.

Therefore, using the 68% interval we may predict that next year’s EPS will grow by some % between 13.1% + 12.1% and 13.1% – 12.1%. This give us a range of 25.2% and 1.0%.

So since last year’s EPS was $1.48 we predict next year’s EPS will be between $1.48 x 1.252 and $1.48 x 1.0100, which is an EPS range between $1.85 and $1.49.

For more insight, Instead of doing this on EPS you can do it on all the factors that combine to give EPS (sales, cogs, sg&a, depreciation, etc.) . You might want to build a pdf that matches the actual historical data too, instead of just assuming everything is normally distributed.

What is the 2nd interesting method by which we can get the range rather than the exact number? Can you talk about it detail?

Thanks
Divakar

There are too many methods of estimating future earnings based on the past to mention – so many different kinds of models! Apart from the one I listed above (least squares fit), another interesting technique is to model the important Income Statement line items related to earnings (sales, cogs, etc.) as normally distributed random variables with a standard deviation based on the past. Doing it that way, you get a range rather than an exact number for what future earnings are likely to be, which might allow you to predict best and worst case, not just the expectation based on average.

Hope that helps,
-lumi

Can you tell what are the methods for estimating future earning by using previous years financial statement.

Some Research contributors are doing estimats based on company’s Guidance, for what extent it is supposable, is it reliable .

Thanks
Gangadhar

If I’m understanding your question correctly, I think one way around this problem is to possibly extrapolate the EPS to the same timeframe as price-per-share so that you have the same number of data points for both for the correlation computation.

For example, say you had prices on a monthly basis, so for one quarter you have 3 prices (perhaps the open or closing price for each month). However you only have 1 EPS for that quarter, so extrapolate this to 3 points by multiplying that single EPS by 33% for the earnings at the end of month #1, 66% for the earnings at the end of month 2, and the EPS number as is for earnings at the end of the 3rd month (end of the quarter).

Or something along those lines. Hope you find something interesting!

Lumilog