. 83
( 208 .)


Prospective Analysis: Forecasting

10-11 Part 2 Business Analysis and Valuation Tools

expenditure assumption. The end result of these assumptions was that the net PP&E turn-
over was forecasted to decline somewhat from the level in 1998. The analyst had to make
assumptions on a few other smaller line items”other operating current assets and lia-
bilities, and other operating long-term assets and liabilities. These assumptions were not
explained by the analyst.
To forecast net debt and equity, the analyst had to assume a ratio of net debt to net
capital (or equivalently, net debt to equity). In 1998 Nordstrom had a ratio of approxi-
mately 35 percent net debt to equity. The analyst assumed that this ratio will increase to
about 40 percent in 1999. The relatively low net debt ratio in 1998 is due to an unusually
large cash balance that Nordstrom built up in 1998. The forecast assumes that Nordstrom
will use that cash to reduce its debt or to buy back stock. Therefore, the 1999 forecasted
balance sheet has 40.4 percent total debt to net capital and 0.7 percent cash to net capital,
and 60.3 percent equity to net capital, whereas the 1998 ratios were 46.8 percent debt to
net capital, 11.9 percent cash to net capital, and 65.1 percent equity to net capital.
The forecasted balance sheet and income statement for Nordstrom imply an increase
in the company™s ROE from 15.6 percent in 1998 to 16.6 percent in 1999. This increase
in ROE is driven by an assumed increase in net pro¬t margin from 4.1 to 4.2 percent, a
decrease in operating asset turnover from 2.49 to 2.37, and an increase in net operating
assets to equity from 1.54 to 1.66 (or equivalently, an increase in net debt to equity from
0.54 to 0.66). These forecasts assume that Nordstrom will continue its recently adopted
strategy of emphasizing pro¬tability and shareholder value.
An alternative approach to balance sheet projection is to assume the change in each
balance sheet account is linked to the change in sales. For example, one might forecast
that inventory balances will increase by 15 to 20 percent of sales increases. The weakness
of this approach is that it takes the beginning balances as given and adjusts from those
points. This is problematic because working capital accounts at a given point in time
often re¬‚ect some unusual deviation from the norm (for example, beginning-of-year
accruals might have ballooned, depending on where payday falls on the calendar). More
important, the ¬rm™s strategy may suggest a shift from the beginning-of-year position.

The Forecast of Cash Flows
The forecast of earnings and balance sheet accounts implies a forecast of cash ¬‚ows.
Table 10-3 shows the projection of cash ¬‚ows for Nordstrom for 1999, using the cash
¬‚ow analysis model discussed in Chapter 9. These forecasts are based on the projected
balance sheet for 1999 and the actual balance sheet for 1998, as shown in Table 10-2.6
The cash ¬‚ow forecasts begin with the projected income for 1998. To this we add
back projected after-tax net interest expense and depreciation to arrive at operating cash
¬‚ow before working capital investments. The forecasted operating cash ¬‚ow before
working capital for 1999 is slightly higher than in 1998. The projected working capital
levels at the end of 1999 imply a net investment of $246 million. Notice how this differs
from a signi¬cant reduction in working capital in 1998. The level of PP&E and other
386 Prospective Analysis: Forecasting

Prospective Analysis: Forecasting

Table 10-3 Analyst™s Forecasts of Nordstrom™s 1999 Cash Flows

1999 Forecast 1998 Actual
$ Millions $ Millions
Net income 238 207
After-tax net interest expense 37 31
Depreciation and other long-term operating accruals 196 187
Operating cash ¬‚ow before investment in
working capital 471 425
Net investment in operating working capital (246) 199
Operating cash ¬‚ow 225 623
Net investment in long-term operating assets and
liabilities (304) (259)
Free cash ¬‚ow available to debt and equity (79) 364
After-tax net interest expense (37) (31)
Net debt (repayment) or issuance 13 258
Free cash ¬‚ow available to equity (103) 591
Cash dividends and repurchase of common stock (123) (375)
Net cash increase (decrease) (202) 216
Ending cash balance (226) 241
Source: Forecasted balance sheet for 1999 from “Nordstrom: Shareholders should be as satis¬ed as customers,”
by B. Missett et. al., Morgan Stanley Dean Witter, December 2, 1998, and the actual balance sheet for 1998
issued by Nordstrom.

operating long-term assets and liabilities implies a net investment of $304 million, lead-
ing to a cash ¬‚ow de¬cit of $79 million. After-tax interest payment was projected to be
37 million dollars. Total debt is projected to be $13 million higher at the end of 1999
relative to the 1998 actual level. Thus, there is a projected 103-million-dollar cash ¬‚ow
de¬cit before dividends and stock repurchases. Despite this de¬cit, Nordstrom is pro-
jected to make a 123-million-dollar payout to equity holders in the form of dividends
and share buybacks because of the large cash balance available at the end of 1998. The
net result is a decrease in cash balance from $241 million to $15 million.

The projections discussed thus far represent nothing more than a “best guess.” Managers
and analysts are typically interested in a broader range of possibilities. For example, in
considering the likelihood that short-term ¬nancing will be necessary, it would be wise
to produce projections based on a more pessimistic view of pro¬t margins and asset turn-
over. Alternatively, an analyst estimating the value of Nordstrom should consider the
sensitivity of the estimate to the key assumptions about sales growth, pro¬t margins, and
Prospective Analysis: Forecasting

10-13 Part 2 Business Analysis and Valuation Tools

asset utilization. What if Nordstrom™s emphasis on pro¬tability results in less sales
growth than anticipated? What if the anticipated improvements in pro¬t margins do not
There is no limit to the number of possible scenarios that can be considered. One sys-
tematic approach to sensitivity analysis is to start with the key assumptions underlying
a set of forecasts and then examine the sensitivity to the assumptions with greatest un-
certainty in a given situation. For example, if a company has experienced a variable pat-
tern of gross margins in the past, it is important to make projections using a range of
margins. Alternatively, if a company has announced a signi¬cant change in its expansion
strategy, asset utilization assumptions might be more uncertain. In determining where to
invest one™s time in performing sensitivity analysis, it is therefore important to consider
historical patterns of performance, changes in industry conditions, and changes in a
company™s competitive strategy.

Seasonality and Interim Forecasts
Thus far, we have concerned ourselves with annual forecasts. However, especially for
security analysts in the U.S., forecasting is very much a quarterly game. Forecasting
quarter by quarter raises a new set of questions. How important is seasonality? What is
a useful point of departure”the most recent quarter™s performance? The comparable
quarter of the prior year? Some combination of the two? How should quarterly data be
used in producing an annual forecast? Does the item-by-item approach to forecasting
used for annual data apply equally well to quarterly data? Full consideration of these
questions lies outside the scope of this chapter, but we can begin to answer some of
Seasonality is a more important phenomenon in sales and earning behavior than one
might guess. It is present for more than just the retail sector ¬rms that bene¬t from hol-
iday sales. Seasonality also results from weather-related phenomena (e.g., for electric
and gas utilities, construction ¬rms, and motorcycle manufacturers), new product intro-
duction patterns (e.g., for the automobile industry), and other factors. Analysis of the
time series behavior of earnings for U.S. ¬rms suggests that at least some seasonality is
present in nearly every major industry.
The implication for forecasting is that one cannot focus only on performance of the
most recent quarter as a point of departure. In fact, the evidence suggests that, in fore-
casting earnings, if one had to choose only one quarter™s performance as a point of de-
parture, it would be the comparable quarter of the prior year, not the most recent quarter.
Note how this ¬nding is consistent with the reports of analysts or the ¬nancial press;
when they discuss a quarterly earnings announcement, it is nearly always evaluated rel-
ative to the performance of the comparable quarter of the prior year, not the most recent
Research has produced models that forecast sales, earnings, or EPS based solely on
prior quarters™ observations. Such models are not used by many analysts, since analysts
388 Prospective Analysis: Forecasting

Prospective Analysis: Forecasting

have access to much more information than such simple models contain. However, the
models are useful for helping those unfamiliar with the behavior earnings data to under-
stand how it tends to evolve through time. Such an understanding can provide useful
general background, a point of departure in forecasting that can be adjusted to re¬‚ect de-
tails not revealed in the history of earnings, or a “reasonableness” check on a detailed
Using Qt to denote earnings (or EPS) for quarter t, and E(Qt) as its expected value,
one model of the earnings process that ¬ts well across a variety of industries is the so-
called Foster model7:
E ( Qt ) Q t -4 + δ + φ ( Q t -1 “ Q t -5 )
Foster shows that a model of the same form also works well with sales data.
The form of the Foster model con¬rms the importance of seasonality because it
shows that the starting point for a forecast for quarter t is the earnings four quarters ago,
Qt-4. It states that, when constrained to using only prior earnings data, a reasonable fore-
cast of earnings for quarter t includes the following elements:
the earnings of the comparable quarter of the prior year (Qt-4);
a long-run trend in year-to-year quarterly earnings increases (δ);
a fraction (φ) of the year-to-year increase in quarterly earnings experienced most
recently (Qt-1 “ Qt-5).
The parameters δ and φ can easily be estimated for a given ¬rm with a simple linear
regression model available in most spreadsheet software.8 For most ¬rms, the parameter
φ tends to be in the range of .25 to .50, indicating that 25 to 50 percent of an increase in
quarterly earnings tends to persist in the form of another increase in the subsequent quar-
ter. The parameter δ re¬‚ects, in part, the average year-to-year change in quarterly earn-
ings over past years, and it varies considerably from ¬rm to ¬rm.
Research indicates that the Foster model produces one-quarter-ahead forecasts that
are off, on average, by $.30 to $.35 per share.9 Such a degree of accuracy stacks up sur-
prisingly well with that of security analysts, who obviously have access to much infor-
mation ignored in the model. As one would expect, most of the evidence supports
analysts™ being more accurate, but the models are good enough to be “in the ball park”
in most circumstances. Thus, while it would certainly be unwise to rely completely on
such a naïve model, an understanding of the typical earnings behavior re¬‚ected by the
model is useful.
Nordstrom™s quarterly EPS for years prior to 1999 behaved as shown in Table10-4.
Note the strong presence of seasonality. The second and fourth quarters of each year
have higher earnings than the other two quarters; the fourth quarter of the year has been
the strongest in every year except 1989, 1991, and 1996.
If we used the Foster model to forecast EPS for the ¬rst quarter of 1999, we would
start with EPS of the comparable quarter of 1998, or $0.215. We would then expect some
additional upward trend in EPS, and a partial repetition of the most recent quarter™s in-
Prospective Analysis: Forecasting

10-15 Part 2 Business Analysis and Valuation Tools

Table 10-4 Nordstrom™s Quarterly Primary EPS, 1988“1998

Fiscal Year EPS Quarter 1 EPS Quarter 2 EPS Quarter 3 EPS Quarter 4
1988 0.120 0.225 0.120 0.290
1989 0.140 0.235 0.135 0.195
1990 0.080 0.220 0.125 0.285
1991 0.155 0.305 0.120 0.250
1992 0.130 0.255 0.145 0.305
1993 0.070 0.260 0.155 0.370
1994 0.195 0.385 0.230 0.425
1995 0.170 0.325 0.180 0.335
1996 0.160 0.275 0.210 0.265
1997 0.205 0.380 0.235 0.380
1998 0.215 0.470 0.270 0.470

crease ($0.470 ’ $0.270). More speci¬cally, when the parameters δ and φ are estimated
with the data in Table 10-4 10, the Foster model predicts EPS of $0.255:
E ( Qt ) Q t -4 + 0.01 + 0.44 ( Q t -1 “ Q t -5 )
E ( Qt ) 0.215 + 0.01 + 0.44 ( 0.470 “ 0.380 ) = 0.255
The model can be extended to forecast earnings two quarters ahead, and even to pro-
duce a forecast for all quarters of the next year. The issue that arises here is that, in fore-
casting earnings two quarters ahead, one needs earnings one quarter ahead, and that
quarter™s earnings are still unknown. The proper resolution of the issue is to substitute
the forecast of next quarter™s earnings. Our forecast of earnings for Nordstrom for the
second quarter of 1999, based on data through the fourth quarter of 1998, would be
E ( Q t +1 ) Q t -3 + 0.01 + 0.44 [ E ( Q t ) “ Q t -4 ]
E ( Q t +1 ) 0.380 + 0.01 + 0.44 ( 0.255 “ 0.215 )
The $0.255 forecast for the ¬rst quarter of 1999, naïve as it is, is not far from the
0.220 actual EPS for Nordstrom in that quarter. Part of the reason that the naïve model
produces a higher forecast is that Nordstrom had an unusually strong fourth quarter in
1998. The model assumes that 44 percent of the EPS increase of the most recent quarter
will carry forward into 1999, but that increase re¬‚ected a one-time effect of the com-
pany™s shift in strategy. The Foster model is not intended as a potential substitute for the
hard work of producing a detailed forecast. Forecasting quarterly earnings should be
done using the same approach used earlier for annual earnings”a line-item by line-item
projection. However, the model does remind us of some important issues. First, it under-
scores that, due to seasonality, a reasonable starting point in quarterly forecasting is usu-
ally the comparable quarter of the prior year, not the most recent quarter. Second, it


. 83
( 208 .)