=Paper=
{{Paper
|id=Vol-1885/86
|storemode=property
|title=Organizational Information
improves Forecast Efficiency of Correction Techniques
|pdfUrl=https://ceur-ws.org/Vol-1885/86.pdf
|volume=Vol-1885
|authors=Florian Knöll,Viliam Simko
|dblpUrl=https://dblp.org/rec/conf/itat/KnollS17
}}
==Organizational Information
improves Forecast Efficiency of Correction Techniques==
https://ceur-ws.org/Vol-1885/86.pdf
J. Hlaváčová (Ed.): ITAT 2017 Proceedings, pp. 86–92
CEUR Workshop Proceedings Vol. 1885, ISSN 1613-0073, c 2017 F. Knöll, V. Simko
Organizational Information improves Forecast Efficiency of Correction
Techniques
Florian Knöll1 and Viliam Simko2
1 Karlsruhe Institute of Technology (KIT),
Fritz-Erler-Straße 23, 76133 Karlsruhe, Germany
knoell@kit.edu
2 FZI Research Center for Information Technology at the Karlsruhe Institute of Technology,
Haid-und-Neu-Str. 10-14, 76131 Karlsruhe, Germany
simko@fzi.de
Abstract: Financial services within corporations have an risks will result in increased costs or uncovered currency
essential need for accurate forecasts. In corporations, ex- risks.
perts typically generate judgmental cash flow forecasts in
a decentralized fashion and provide data that is import- 1.1 The Problem of Judgmental Forecasts
ant in corporate risk management. But the accuracy of
these forecasts is most likely reduced by biases of the or- Usually, cash flow forecasts result from the judgment of
ganizational structure. As for the importance of cash flow human experts [24] and are revised several months or quar-
forecasts, usually correction techniques are applied with ters after the initial forecast until the date of the actual
statistical methods based on historical data. In most cases realization finalizes the sequence of forecasts. The initial
the organizational biases are not included into the correc- forecast and the sequence of adjusted forecasts is referred
tion techniques. This paper argues that disregarding the to as forecasting process, while the sequence of adjust-
organizational information actually decreases forecast effi- ments in revisions is usually coined as revisioning pro-
ciency. Forecast efficiency provides statistical information cess or simply revisioning. When judgmental forecasting
for the amount of structure within forecasts and errors. In takes place, the forecasts can be prone to individual bi-
case of aggregated cash flows in accounting, the forecasts ases and latent human factors that entail forecasting pro-
highly depend on return margins. The empirical results cesses in many ways [16, 18]. Additionally, the organiza-
in this paper show that debiasing with forecasts correction tional structures and dependencies of the environment can
based on organizational information can improve forecast change the forecaster’s expectation, resulting in organiza-
efficiency by 56 % to a statistical approach. The reduc- tional biases that result in forecast inaccuracies [6].
tion of inefficient pattern show statistics arguing for fore-
cast correction that rely on organizational biases (stand- 1.2 Correction Techniques and Organizational Biases
ard deviation of error 0.20) instead of basic statistical ap- Improving biased forecasts is possible with forecast cor-
proaches that harm forecast efficiency (standard deviation rection techniques that analyze and change the human pre-
of error 0.28). diction with statistical models [12]. For instance, [15]
found dependencies of timing and magnitude of cash flow
1 Introduction revisions. Their results state that cash flow forecast pro-
cesses are more accurate when they show a high revision
Corporations with global operations typically generate at a late state of the process compared to a high revision at
forecasts for cash flow items on a regular basis (e.g., the early stage.
monthly or quarterly), at different organizational levels, However, current forecast correction techniques often
business divisions, and countries. These forecasts are of- employ solely statistical methods – leaving out the organ-
ten generated in a decentralized fashion by the subsidiar- izational biases for approaches of forecast improvement.
ies, where the subsidiaries send thousands of item-level In corporate finance, several important key performance
forecasts and revisions to corporate headquarters. These indicators (KPI) exist that aggregate many figures. An
forecasts are then consolidated and used in crucial tasks of example of such key figure is Earnings Before Interest,
the corporate finance department (such as in [14] or even to Taxes, Depreciation, and Amortization (EBITDA) margin,
access with cash flow forecasts the company’s stock mar- which can be used as one of the primary proxies for a
ket value [13]). company’s current operating profitability [19]. When hu-
The tasks in corporate departments strongly depend on mans try to achieve personal objectives (e.g., bonus pay-
the quality of the forecasts, as they provide the data base ments by financial incentives) predefined targets that rely
for the financial planning operations and subsequent man- on these figures, for instance percentage return margins,
agement activities. For instance, due to forecast inac- these organizational biases can alter forecasts and their ad-
curacies, the corporate hedging to reduce foreign exchange justments in a revisioning process [11].
Organizational Information improves Forecast Efficiency of Correction Techniques 87
In addition, in the realm of cash flows, several business 2 Related Work
functions might influence the realization volume of cash
flows. The looming failure to meet earnings targets (which Organizational biases can result in forecast inaccuracies as
might reduce manager’s bonus payments) is an incentive pointed out by Daniel et al. [6] but does not correct them
to hold-back invoices received within term of credit. Al- in any way. He identified "dividend thresholds" as a or-
ternatively, managers can trigger invoices issued earlier or ganizational bias, which alters the forecasts.
might change payment terms in order to align annual cash In the paper [21], the authors analyzed short time series
results with targets. Conversely, if earning targets have within the year and used a Bayesian method to account
been met already, there might be an incentive to delay the for sub-seasonal information for the seasonal based cor-
issuing of invoices until the next year to increase the prob- rection. In contrast to their setting, our forecast series are
ability of meeting next year’s targets. In particular, the even shorter (5 reference points instead of 12), the applica-
papers of [4], [7], and [5] show that realizations are often tion of linear regression models (instead of Bayesian mod-
shifted according to earnings management policies. When els), and we account for one single information in our pa-
the volumes are shifted, the forecast errors can be expected per focuses a margin target at the end of year (instead of
to exhibit a systematic bias. the whole sub-annual pattern).
Regarding seasonality, Yelland [27] concludes that a
1.3 Efficiency Theory simple stable seasonal pattern model can perform surpris-
Biases often translate to observable patterns in forecast- ingly well, if it uses “theory-free” descriptions of booking
ing processes and one measurement to analyze the sys- processes. His findings are in resonance to the theme that
tematic behavior of revisioning is the efficiency theory. simple empirically-based models do frequently better than
The theory in market finance [10] and forecasting [22] complex ones.
suggest that processes are efficient if they describe a ran- The authors of [3] promote that in marketing and fin-
dom walk. The theory states that non-random walks ance simple models sometimes predict more accurately
promote inefficient forecasting since correlations among than complex models. The authors argue that “the benefits
revisions with revisions or errors are expected to show of simplicity are often overlooked because the importance
statistical insufficiency that has the potential to anticip- of the bias component of prediction error is inflated, and
ate future adjustments or errors. The application of this the variance component of prediction error (based on over-
theory provides evidence that correlations exist in many sensitivity to different samples) is neglected.” Reasoned
cases [2, 17, 1, 9, 8]. by their study, we correct the forecasts with a simple lin-
ear regression model.
1.4 Our Contribution
This paper argues that efficiency provides a statistical tool 3 Empirical Cash Flow Data
to evaluate different correction approaches. The analysis
of efficiency figures can provide insights for the differ- The data stems from a record of cash flow forecasts and
ences of model predictions. The analyses for accounting realizations provided by a multinational sample corpora-
cash flows contribute to the current research as they show tion. With over 100,000 employees, the company gener-
that including organizational information into correction ates annual revenues in the billion Euro range. The cor-
models is key for further improvements in correction tech- poration is headquartered in Germany, but has worldwide
niques. When the empirical outcomes of these organiza- more than 300 separate legal entities. The subsidiaries are
tional models are compared to purely statistical model ap- grouped into four distinct divisions (D1 – D4), based on
proaches they show that both models reduce the error, but their business portfolios.
the disregard of organizational information in the purely Each subsidiary operates officially independently of the
statistical approach does crucially harm the forecast effi- corporation, while there are some organizational depend-
ciency. Moreover, this insight is also applicable to other encies. First, based on the set of local plans, the corpor-
domains, where exploratory data analysis and forecast cor- ation re-adjusts the planning to an overall view, and sets
rection play an important role in time series forecasting. the target requirements for local operations for being rated
as a “successful” subsidiary. Second, in the corporation
1.5 Structure of the Paper the fiscal year ends in December and the subsidiaries that
The remainder of the paper is structured as follows. The meet targets is assumed to be most pronounced at the end
data description in Section 3 is followed by the notation of the year. Third, as the subsidiaries operate independ-
that is introduced in Section 4. Section 5 describes the ently, they have their own financial information system,
design for the empirical analysis and the concept of fore- a heterogeneous payment structure (e.g., incentivization
cast efficiency in detail. Section 6 presents the results and bonuses) and have to ensure liquidity for their operations
interpretation of the analysis. In Section 7 discusses the (e.g., with earnings management processes). Fourth, each
implications of this work for future improvements in fore- subsidiary that is participating in the forecasting process
cast correction. – mostly large-volume entities – enters its expectations on
88 F. Knöll, V. Simko
future cash flow in a digital, corporate-based forecasting S
entity=E
system. (E) t Ry=Y,m=M − min( R)
Financial risk management is centralized, with the local t Ry=Y,m=M = S S
max( R) − min( R)
subsidiaries reporting cash flows to the corporation’s cent- while:
ral finance department, where these serve as the basis for [
further actions in corporate finance. Therefore, the corpor- R = {t Rentity
date : entity = E ∧ date < (Y, M)}
ate finance department receives cash flow forecasts (fore-
casts) generated by the subsidiaries worldwide, denomin- Definition 2 (Target ratio). The suggested annual return
ated in foreign currencies. After the realization date, the target (target ratio) that an entity has to reach at the end of
corporation receives in every month the cash flow figures the year y = Y is defined as:
for realizations (actuals). The data available cover item- T (0 Ry=Y )
types of invoices issued (II) and invoices received (IR)
from the corporate IT system. In order to evaluate possible As targets are unknown (to us), but business develop-
strategies and provide further information for KPI figures ment measured with EBITDA figures seem rather stable
such as percentage return ratio the forecasts and actuals over the years, the target ratio in y = Y is estimated by av-
are aggregated for the corporate risk management. As a eraging the December actual ratios of the three preceding
proxy for the percentage return margin within a fiscal year, years (0 Ry=Y − j,m=12 , for j ∈ {1, 2, 3}).
the entity’s ratio of aggregated revenues (II) and expenses
Definition 3 (Revision). The revision for ratios describes
(IR) is calculated.
the adjustment from the second to last forecast before the
The aggregated data set used in the analysis for this pa-
actual. It is formally defined as;
per covers forecasts and actual for the entity’s ratios. De-
livered by the subsidiaries on a quarterly basis, the fore- 12 R = 1 R − 2 R
casts cover intervals with horizons of up to 15 months
(five quarters). The dataset for actual invoices ranges from This paper uses the last revision because generally the
January 2008 to December 2013 with the corresponding latest judgmental forecast incorporates the most informa-
forecasts covering the actuals’ period. tion and is the most accurate [20].
In total, actuals and forecasts are available for the 67 Definition 4 (Difference from target). The difference from
largest subsidiaries resulting in 25 different currencies for target is defined as:
the dataset. Actuals grouped by division, subsidiary, cur-
rency and item-type result in 72 actual time series. Over- TargetDiff = T (0 R) − 1 R
all, the dataset consists of 3,087 monthly invoice actuals,
Definition 5 (Error). Finally, the error is defined as:
with five associated forecasts each. The underlying raw
dataset of non-aggregated forecasts cover 102.360 items. t E = 0R − t R
Table 1 gives a brief summary of the dataset.
Table 2 gives a brief overview of the defined metrics.
4 Notation and Forecasting Process
5 Research Design
The notation presented in this section is commonly used
Improving forecast accuracy is an important goal, where
in current literature on [22].
usually correction techniques such as linear regressions
Denoting the actual of cash flow margin ratio as 0 R, the
are applied in the literature for analysis and correction
lead time t > 0 of a forecast t R for 0 R refers to a quarter of
of biases. These statistical forecast correction techniques
the year until the actual date (t = 0). Figure 1 visualizes
build models that usually employ information of basic fea-
the temporal structure of an example forecasting process
tures based on historical data. An example of such a basic
in five steps for an actual 0 R. The initial forecast ratio 5 R
statistic model can be found in Def 6. Here, the forecast
is delivered with a lead time of five periods and is revised
error 1 E is regressed using basic variables such as regres-
four times until the last one–period–ahead forecast 1 R is
sion intercept, ratio 1 R, and revision 12 R. Theoretically
generated.
valid, this model optimizes the error based on the human
Since ratios are specific for an entity, for reasons of
forecaster’s prediction and revisioning behavior. But, this
comparability, this work focuses on normalized ratios
paper argues that correction approaches should incorpor-
(Def. 1). Therefore, the notation t R refers to the normal-
ate important organizational information too. As noted be-
ized ratio instead of the entity specific ratio (t R := t R(E) ).
fore, reaching predefined target KPIs is an important stra-
Definition 1 (Normalized ratio). Normalized ratio is ob- tegic goal. The difference to the percentage return margin
tained by subtracting the minimum ratio within an entity target is symbolized with TargetDiff and measures the dis-
from R and dividing by the difference of its maximum and tance to the organizational prerequisites. To overcome this
minimum ratio. The values are always between zero and organizational bias, the information of TargetDiff is integ-
one per entity. rated into the regression model as shown in Def 7.
Organizational Information improves Forecast Efficiency of Correction Techniques 89
Table 1: The summary of the analyzed cash flow data.
Divisions Subsidiaries Currencies Time Series Actuals Forecasts
D1 10 7 11 618 3090
D2 13 8 15 608 3040
D3 6 4 7 420 2100
D4 38 20 39 1441 7205
All 67 25 72 3087 15435
1
Cash Flow Ratio!
5R! 4R ! 3R! 2R! 1R! 0R !
0,75
Forecast!
Revision! Actual!
0,5
0,25
Error!
0
5 4 3 2 1 0!
Time to Present (t)!
Figure 1: Temporal structure of margin ratio forecasts t R (t > 0) with the corresponding actual margin ratio 0 R.
Table 2: Notation used in the analyses. are considered as weak-form efficient. Otherwise, exist-
ing structures hint to information that could be incorpor-
Notation Metric ated into revisions because revisions are predictable. With
tR Forecast Ratio (normalized) t ∈ R+0 denoting the lead-time to the realization of an ac-
tual (at t = 0), Nordhaus suggests testing for weak-form
TargetDiff Difference from target
efficiency using the Propositions (P1) and (P2).
0R Actual Ratio (normalized)
Proposition 1 (P1). Forecast error at t is independent of
12 R Revision
all revisions up to (t + 1).
T (0 R) Target
Proposition 2 (P2). Forecast revision at t is independent
tE Error
of all revisions up to (t + 1).
Combining the argumentation for organizational debi-
Definition 6 (Basic statistic model MBasic ). asing and efficiency, the authors propose the following hy-
potheses:
1 E ∼ β0 + β1 (1 R) + β2 (12 R)
Hypothesis 1. Does forecast correction that incorporates
Definition 7 (Organizational model MOrga ). organizational information (that organizationally biases
forecasts) improve forecast efficiency?
1 E ∼ β0 + β1 (1 R) + β2 (12 R) + β3 (TargetDiff )
Hypothesis 2. How does efficiency for organizational cor-
Typically, correction techniques evaluate their results rection differ from basic statistical approaches?
with some error metric, such as error (deviation), abso-
lute error, percentage error, absolute percentage error, and These hypotheses are evaluated based on the two re-
so on. Slightly different use cases can favor a specific er- gression models. Both models are trained for each month
ror measure as most of them have known flaws that suit of the year independently to consider the seasonality in
one case but not the other ones. The research presented in the business data. Therefore, the data is split into 12 sub-
this paper tries to be independent of those restrictions that sets that are accessed to train one specific model for each
make comparison of scientific results difficult and hinders month (resulting in 24 models). To show the benefit of
reproducibility. Therefore, the comparison of both models the organizational information empirically, the model pre-
is evaluated in an error-metric-independent way. diction needs to add the original forecast 1 R to derive a
Based on the efficiency theory [22], proposed tests for new model prediction. These model predictions will then
the structure in terms of correlations amongst revisions be compared to the original forecasts (M∅ symbolizes the
and between revisions and errors. Forecast processes that expert forecast) and with each other in terms of forecast ef-
show no correlation structures (with significant p-values) ficiency. The baseline for comparison is the original fore-
90 F. Knöll, V. Simko
4E
cast based on M∅ , which will be evaluated first. For reas-
ons of clarity, the model forecast substitutes the original 4E 0
3E
forecast, which leads to three possible forecast processes
“5 R, 4 R, 3 R, 2 R, 1 R(M{∅,Orga,Basic} ), 0 R” with changed re- 3E x 0
2E
vision and error measures for 12 R and 1 E depending on
2E x x
the selected model. Logically, the evaluation focuses on 0
1E
these changed measurements only. Additionally, the in- 1E x x x 0
45R
dication for error quantiles and statistics for efficiency are
provided. 45R 0 0 0 −0.9 0
34R
34R x 0 0 −0.19 0 0
23R
6 Empirical Analysis
23R x x 0 0.7 0 0 0
12R
This section presents the empirical results. These consist
of correlation analysis for efficiency, with a revision and 12R x x x 0.92 −0.72 −0.85 −0.83 0
error analysis, followed by the analysis of the underly-
ing statistics. For the correlation analysis the experiments Figure 3: Shows percentage improvement in correlation
use the R programming language [23] and the libraries of a basic statistical model MBasic over the baseline model
corrplot [25] and knitr [26]. M∅ (positive numbers exemplify the improvement).
As noted before, the forecast efficiency is an important
goal of forecasting processes. The forecast efficiency of
4E
the resulting prediction of the models MOrga and MBasic
4E 0
are compared to each other and the baseline M∅ . The
3E
baseline of forecast efficiency for M∅ is shown in Figure 2. 3E x 0
2E
It should be noted that in the figures, we hide irrelevant
cells (marked using "x" sign) and we show all and only 2E x x 0
1E
the cells relevant for the efficiency analysis as proposed in
[22]. 1E x x x 0
45R
45R
4E
0 0 0 0.73 0
34R
4E
34R
3E
x 0 0 0.16 0 0
23R
3E x
23R
2E
x x 0 0.63 0 0 0
12R
2E x x
12R
1E
x x x 0.56 0.57 0.06 0.19 0
1E x x x
45R
Figure 4: Shows percentage improvement in correlation of
45R our organizational model MOrga over the baseline model
34R
MBasic (positive numbers exemplify the improvement).
34R x
23R
23R x x
12R
in red). Comparison between the basic statistical model
12R x x x and the organizational model in Figure 4 shows an ad-
ditional increase of efficiency relative to MBasic by 56%
for the final forecast. More remakable, the whole fore-
−1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1
casting process is more efficient (see (12 R,23 R), (12 R,34 R)
Figure 2: Shows correlation of revisions with errors and (12 R,45 R)) stating that the organizational debiasing
and revisions of experts, without any correction (baseline approach is superior to the basic statistical approaches.
model M∅ ). The Figure 5 shows important information for the error
quantiles of the forecasts. This figure also provides addi-
The comparison of models M∅ − MBasic and MBasic − tional support for the performance of MOrga through the
MOrga (difference in correlation) are depicted in Figure 3 1 E measure. The organizational model outperforms the
and Figure 4 respectively. statistical model especially for the 1. quartile (∆ = 0.072),
The Figure 3 shows that the basic statistical model median (∆ = 0.017), and 3. quartile (∆ = 0.120). Only for
increases efficiency (marked in blue) compared to the minimum, maximum, and for mean error (∆ = 0.002) the
baseline by (12 R,1 E) = 92% and (23 R,1 E) = 70%. But, all statistical model seems beneficial.
the other dependencies have decreased efficiency (marked The results for Cor(12 R,1 E) are not significant after cor-
Organizational Information improves Forecast Efficiency of Correction Techniques 91
1.0
This study contributes with the conclusion that the dif-
ferent results for corrective models may be inherent to
each approach.
0.5
Relevance for the Data Mining Community
0.0
For the data mining community the paper might change
the understanding of the link between exploratory data
−0.5
analysis and forecast correction. Exploring data can ac-
tually show the way how to correct forecasts in a model-
independent way. We would like to stress that the results
−1.0
of this paper were not achieved with a neural network, a
M0 MBasic MOrga random forest, or a complex machine learning algorithm.
Instead, the results are achieved with a simple linear re-
Figure 5: Quantiles of error 1 E of the expert and the stat-
gression models.
istically / organizationally corrected forecasts.
The importance of exploratory data analysis is
strengthened as data understanding additionally allows a
rection due to the high efficiency, but the details are shown differentiation between biases with pattern and errors.
in Table 3. The Spearman covariance for the approaches The most important result of this study is probably the
states that revisions and error have a lower joint variabil- statement that a basic statistical model “just” tries to op-
ity. The organizational model has a positive covariance, timize the selected component (e.g, the error), while an
while the statistical model has a negative covariance with organizational model tries to reduce the bias itself. As a
a higher magnitude. Also, the table shows that organiza- result of the organizational model enables the possibility
tional model increases standard deviation for the revision, to identify further unknown biases and correct these bi-
but it reduces for the error. It is arguable with these num- ases with a second model. Understanding the error com-
bers that the organizational model’s revision focuses with ponents is important. When a forecaster distinguishes the
meaningful revisions on the reduction of the error, while signal from the noise, the error should decrease by the way
the statistical model’s revision focuses on changing the er- or making predictions more confident. Therefore, even if
ror with minor corrections. This enables future approaches no error decrease is achieved with one organizational debi-
to detect other, currently unknown biases to be identified asing model, a patch of models for the most important or-
and removed. ganizational biases will definitely increase the accuracy.
Overall, the results state several advantages of the or-
ganizational model in comparison to the statistical model.
First, in the sense of Nordhaus the organizational debias- Managerial Implications
ing model improves forecast efficiency for Cor(12 R,1 E), From the perspective of a manager and forecast researcher
supporting Hypothesis 1. Second, the error distribution is it is important to understand in which way business-related
narrowed, especially for the 1st and 3rd Quartile. Third, factors may affect forecasts and indirectly correction mod-
the advantage of bias reduction instead of error optimiza- els. In the case of cash flow forecasts in a corporate setting
tion. The second and third finding support Hypothesis 2. one important factors is the percentage margin target, as
these might provide incentivization to alter forecasts and
7 Conclusions and Outlook actuals of cash flows. The underlying value of this inform-
ation is stated in terms of forecast efficiency. The analysis
Empirical analyses on forecast efficiency or on cash flow showed that efficiency increases.
biases might be a very interesting paper topic for the Based on this research, application of the presented ap-
specific research communities and therefore easy to find. proach would be interesting also for forecasting in other
However, linking these settings to forecast correction tech- domains. The efficiency theory could provide an alternat-
niques that account for organizational biases in a predict- ive approach to understand the value of specific informa-
ive model have not been explored in the forecast com- tion within forecast correction (compared to other meas-
munity so far. ures such as entropy or information gain).
This research addresses two research gaps: (1) Link-
ing organizational information to forecast correction tech- Outlook
niques and evaluating the result independently from a spe-
cific error metric. The results show that organizational It might be reasonable to recommend in the forecasting
information is beneficial to forecast efficiency. (2) Ana- community that future approaches shall not minimize the
lyses of correction models that compare basic statistical error component, by changing forecasts and revisions mar-
approaches to organizational approaches have been left ginally. Instead, maximization or at least the change of
unattended. forecasts and revisions in an acceptable big magnitude that
92 F. Knöll, V. Simko
Approach Covariance(12 R,1 E) Std.Dev.(12 R) Std.Dev.(1 E)
M∅ (Baseline) -246092.58 0.24 0.34
MOrga (Organizational) 8968.19 0.28 0.20
MBasic (Statistical) -20360.87 0.21 0.28
Table 3: Table shows metric details for Spearman correlation values of the revision and the error in ratio of the expert and
the organizationally / statistically corrected forecasts.
result in marginally errors is recommended. A high revi- [13] S. N. Kaplan and R. S. Ruback. The Valuation of Cash
sion will determine how long the forecast result is aligned Flow Forecasts: An Empirical Analysis. The Journal of
to the bias pattern. Based on the results, the understand- Finance, 50(4):1059–1093, 1995.
ing of forecasts and best applied correction techniques is [14] C. S. Kim, D. C. Mauer, and A. E. Sherman. The Determ-
obtained on the way. inants of Corporate Liquidity: Theory and Evidence. Jour.
of Financial and Quant. Analysis, 33(3):335–359, 1998.
[15] Florian Knöll, Verena Dorner, and Thomas Setzer. Relating
Cash Flow Forecast Errors to Revision Patterns. In Proc. of
References MKWI, pages 1217–1228. MKWI - Prescriptive Analytics
in IS, Universitätsverlag Ilmenau, 2016.
[16] Michael Lawrence, Paul Goodwin, Marcus O’Connor, and
[1] Masahiro Ashiya. Testing the Rationality of Forecast Re- Dilek Önkal. Judgmental Forecasting: A Review of Pro-
visions Made by the IMF and the OECD. Journal of Fore- gress Over the Last 25 Years. International Journal of Fore-
casting, 25(1):25–36, 2006. casting, 22:493–618, 2006.
[2] D. A. Bessler and J. A. Brandt. An Analysis of Forecasts [17] Michael Lawrence and Marcus O’Connor. Sales Forecast-
of Livestock Prices. Journal of Economic Behavior & Or- ing Updates: How Good are They in Practice? Interna-
ganization, 18(2):249–263, 1992. tional Journal of Forecasting, 16(3):369–382, 2000.
[3] Henry Brighton and Gerd Gigerenzer. The Bias Bias. [18] J. Leitner and U. Leopold-Wildburger. Experiments on
Journal of Business Research, 68(8):1772–1784, 2015. Forecasting Behavior with Several Sources of Information–
[4] David Burgstahler and Ilia Dichev. Earnings Management A Review of the Literature. European Journal of Opera-
to Avoid Earnings Decreases and Losses. Journal of Ac- tional Research, 213(3):459–469, 2011.
counting and Economics, 24(1):99–126, 1997. [19] B. Marr. Key Performance Indicators (KPI): The 75 Meas-
[5] David Burgstahler and Michael Eames. Management of ures Every Manager Needs to Know. Pearson UK, 2012.
Earnings and Analysts’ Forecasts to Achieve Zero and [20] S. K. McNees. The Role of Judgment in Macroeconomic
Small Positive Earnings Surprises. Journal of Business Fin- Forecasting Accuracy. International Journal of Forecast-
ance & Accounting, 33(5–6):633–652, 2006. ing, 6(3):287–299, 1990.
[6] N. D. Daniel, D. J. Denis, and L. Naveen. Do Firms Man- [21] M. Mendoza and E. de Alba. Forecasting an accumulated
age Earnings to Meet Dividend Thresholds? Journal of series based on partial accumulation ii: A new bayesian
Accounting and Economics, 45(1):2–26, 2008. method for short series with stable seasonal patterns. Inter-
[7] Francois Degeorge, Jayendu Patel, and Richard Zeck- national Journal of Forecasting, 22(4):781–798, 2006.
hauser. Earnings Management to Exceed Thresholds. [22] W. D. Nordhaus. Forecasting Efficiency: Concepts and
Journal of Business, 72(1):1–33, 1999. Applications. The Review of Economics and Statistics,
[8] Bruno Deschamps and Christos Ioannidis. Can Rational 69(4):667–674, 1987.
Stubbornness Explain Forecast Biases? Journal of Eco- [23] R Core Team. R: A Language and Environment for Statist-
nomic Behavior & Organization, 92:141–151, 2013. ical Computing. Vienna, Austria: R Foundation for Statist-
[9] Jonas Dovern and Johannes Weisser. Accuracy, Unbiased- ical Computing., 2013.
ness and Efficiency of Professional Macroeconomic Fore- [24] N. R. Sanders and K. B. Manrodt. The Efficacy of Us-
casts: An Empirical Comparison for the G7. International ing Judgmental Versus Quantitative Forecasting Methods
Journal of Forecasting, 27(2):452–465, 2011. in Practice. Omega, 31(6):511–522, 2003.
[10] E. F. Fama. Efficient Capital Markets: A Review of Theory [25] Taiyun Wei and Viliam Simko. corrplot: Visualization of a
and Empirical Work. The Journal of Finance, 25(2):383– Correlation Matrix, 2016. R package version 0.77.
417, 1970.
[26] Yihui Xie. knitr: A Comprehensive Tool for Reproducible
[11] F. Guidry, A. J. Leone, and S. Rock. Earnings-Based Bonus Research in R. Chapman and Hall/CRC, 2014. ISBN 978-
Plans and Earnings Management by Business-Unit Man- 1466561595.
agers. Journal of Accounting and Economics, 26(1):113–
[27] P. M. Yelland. Stable Seasonal Pattern Models for Forecast
142, 1999.
Revision: A Comparative Study. International Journal of
[12] Jiawei Han, Jian Pei, and Micheline Kamber. Data Mining: Forecasting, 22(4):799–818, 2006.
Concepts and Techniques. Elsevier, 2011.