Cjss second Issue: cjss second Issue qxd

Problem
The study on which this paper is based investigated the problem of fore-
casting under conditions of a high degree of uncertainty (Raupp & Apkhazava,
2009). Specifically, that study aimed to forecast tourism in post-Soviet Geor-
gia, acknowledging the uncertainty caused by a volatile global economy, bel-
ligerent domestic opposition, and threat of recurring armed invasion from
the Russian Federation.
Background
Forecasting under the best of conditions is characterized by uncertainty.
Traditional forecasting methods make a critical assumption, i.e., that condi-
tions in the near future will look very much like conditions in the recent past.
All other things being equal, the assumption may be useful for the task of fore-
casting in the relatively near future, perhaps up to a few years. However, it is
seldom the case that all things in the future will look like all things in the past.
It is necessary, therefore, to use non-traditional methods to forecast such
highly uncertain variables as tourism in post-Soviet Georgia.
Methods
Recognizing the weakness of traditional methods alone in forecasting
under conditions of a high degree of uncertainty, Raupp and Apkhazava
(2009) used a mix of statistical and judgmental methods. Each of these is de-
scribed below.
Statistical Methods
Causal models relate the variable to be forecasted to one or more predic-
tor variables. Perhaps the best known is what economists call “the law of de-
mand” and “the law of supply.” As the price of a good rises the quantity
demanded of that good falls; and as the price of a good rises, the quantity sup-
plied of that good also rises. Each of these is a simple regression model. We
know, however, that other factors determine the quantities of a good de-
manded or supplied, e.g., as income rises the quantity demanded also rises,
and as a drought produces lower yields, the quantity supplied falls. These are
examples of a multiple regression model, and each asserts a causal relation-
ship.
31
Caucasus Journal of Social Sciences

---

Non-causal models make no claims that changes in the variable to be
forecasted are known to be caused by changes in any other variable. The most
common non-causal model is time series. In this case, we do not know from
the data why the variable changes, but we discern several elements in the se-
ries:  trend, seasonality, cycles, and unexplained factors. One of the most ac-
curate computer programs available on the market today to deal with time
series is ForecastPro (Stellwagen & Goodrich, 2008), which is used exten-
sively by T
HE
G
EORGIA
F
ORECAST
TM
and The University of Georgia Forecasting
Center.
Judgmental Methods
In seeking the most accurate forecast, it is sometimes helpful to adjust
the statistical forecast using one or more judgmental methods. Perhaps the
forecaster knows something about the past or the future that may not be re-
flected in the data.
Analogies include examples of past situations that may be similar in key
respects to the situation today. The tourism study used the Bali bombings,
Assisi earthquakes, and Kauai hurricane Iniki to gain insights into what hap-
pens to tourism after a disaster. The study concluded that tourism falls
sharply in the short term but may recover given investments in infrastructure
and marketing. 
Delphi uses a panel of experts to collect data in areas where clear statis-
tical evidence is not available (Armstrong, 1999). Experts provide their judg-
ments anonymously, receive feedback on what other experts are saying, and
have a chance in a second or third round to adjust their estimates.
Surveys are useful to collect data on opinions in the surveyed population.
They have been used by T
HE
G
EORGIA
F
ORECAST
TM
(Raupp, 2009) since 2007 to
calculate a Consumer Confidence Index and a Producer Confidence Index.
Surveys are conducted in Gori, Tbilisi, Batumi, and Kutaisi.
Game theory is a powerful tool in decision-making when there are only
two or a few actors, e.g., in oligopolies. This method was not used in the
tourism study, but it is used by T
HE
G
EORGIA
F
ORECAST
TM
in other studies.
Combining Methods
Given results of statistical forecasts and various judgmental methods,
the forecaster may decide to combine the forecasts. Combining forecasts is a
subject of considerable discussion and research among forecasting profes-
sionals (e.g., see Armstrong, 1989). When forecasters adjust statistical fore-
32
Caucasus Journal of Social Sciences

---

casts by using judgmental factors, the must explain in detail what was done
and why, and they must document the process. A set of rules should be es-
tablished in advance. 
The tourism study uses a “6-8-10” rule with respect to its monthly na-
tion-wide confidence survey. There are two classes of subjects (consumers
and producers) in each of four cities, resulting in a 2-by-4 matrix of 8 cells. If
6 of the 8 cells are in agreement (either above or below 1.0), then adjust the
statistical forecast by 10 percent; if, however, the trend in those 6 cells is op-
posite of the ratio, make no adjustment, as the results are ambiguous.
Scenario scripting is the final step in creating forecasts under conditions
of a high degree of uncertainty. The reality is that a single point forecast is
not feasible. We simply do not know what will happen to the global economy.
Nor do we know what civil disturbances may break out. And we cannot know
the mind of Vladimir Putin. Therefore, if we cannot predict, then we can pre-
pare (see Taleb, 2007). 
The tourism study scripted nine scenarios, as shown below. Three basic
scenarios are Best Case (B), Middle Case (status quo) (M), and Worst Case
(W). Each was analyzed using three forces: Macroeconomic Forces (V1), Do-
mestic Forces (V
2
, and Russian Forces (V
3
). An expected value matrix was cal-
culated based on expert judgment. In the Best Case scenario, Georgia could
expect a substantial increase in tourism, while in the Worst Case scenario,
tourism would essentially vanish.
Figure 1. Summary of scenario scripts.
33
Caucasus Journal of Social Sciences
Best Case (B)
Middle Case (M)
Worst Case (W)
Macroeconomic Forces
(V
1
)
B
1
More tourists can
afford to travel to
Georgia. Improved
infrastructure soon.
M
1
Some tourists can
afford to travel to
Georgia. Improved
infrastructure later.
W
1
Few tourists can afford
to travel to Georgia.
Little work on
infrastructure.
Domestic Forces (V
2
)
B
2
Tourists are not put off
by civil unrest.
M
2
Some tourists are
discouraged by civil
unrest.
W
2
Most tourists are
discouraged by civil
unrest.
Russian Forces (V
3
)
B
3
Tourists are not afraid
of armed conflict.
M
3
Some tourists fear
Russian invasion.
W
3
Tourists do not travel
to war-torn Georgia.

---