Bio-based and Applied Economics 5(1): 63-81, 2016

ISSN 2280-6180 (print) © Firenze University Press 
ISSN 2280-6172 (online) www.fupress.com/bae

Full Research Article

DOI: 10.13128/BAE-16367

The instability of farm income. Empirical evidences on 
aggregation bias and heterogeneity among farm groups

Simone Severini*, AntonellA tAntAri, GiuliAno Di tommASo

Università della Tuscia, DAFNE, Viterbo, Italy

Date of submission: 2015 29th, June; accepted 2016 16th, February

Abstract. This paper analyses the instability of farm income experienced by a con-
stant sample of Italian farms over the period 2003-2012. It assesses the extent of the 
aggregation bias due to the use of aggregated vs. single farm data and estimates the 
level of farm income variability in several groups of farms for the whole period and 
for two sub-periods. Differences between groups and periods are assessed by means of 
non-parametric tests. 
Results suggest that analyses based on aggregated farm data are likely going to stron-
gly underestimate the extent of income variability faced by farmers. Income variability 
levels differ among farm groups and have significantly increased over the considered 
time. This has policy implications regarding the risk management tools recently intro-
duced within the Rural Development Policies and how these should be targeted on the 
farms that more need them.

Keywords. Farm income, income instability and variability, aggregation bias, farm 
heterogeneity, risk management policies

JEL codes. Q12, G320, Q18

1. Introduction

Farming is a risky business because forces (such as weather) beyond the control of 
farmers affect their income (Mishra and Sandretto, 2002). Farm income stability has been 
one of the goals of agricultural policies both in the US and the EU (Mishra and Sandretto, 
2002: 209). This is because income instability negatively affects farmers’ well-being and deci-
sions, their ability to expand operations and repay debt and, in turns, this can also have sec-
ondary effects on agribusiness firms and creditors (Mishra and El-Osta, 2001; Mishra and 
Holtausen, 2002; Mishra and Sandretto, 2002; Vrolijk and Poppe, 2008; Vrolijk et al., 2009).

While a large number of studies focuses on price and/or yield and revenue instabil-
ity (looking often at single crops or very specialised farms), there are not many analyses 
specifically focused on the stability of the whole farm income. This seems an important 

* Corresponding author: severini@unitus.it



64 S. Severini, A. Tantari, G. Di Tommaso

knowledge gap because “… farmers are ultimately concerned more about their net incomes 
than about prices and costs” (Mishra and Sandretto, 2002: 219). The lack of empirical 
evidences on income instability in the EU may become a constraint for monitoring and 
designing the set of tools that have been introduced by the recently reformed Common 
Agricultural Policy (CAP) to support farmers to cope with risk (Matthews, 2010; Meuwis-
sen et al., 2011; Tangermann, 2010)1.

This paper tries to fill this gap investigating the level of income variability of a large 
constant sample of Italian farms over the period 2003-2012. This allows to assess the level 
of farm income variability in the whole considered sample and in several farms grouped 
according to production orientation, size and economic performances as well as of two 
consecutive periods of time. 

The paper aims at identifying the information that are useful for designing and tar-
geting income stability policy tools focusing on two aspects. On the one hand, it assesses 
whether it is preferable to use variability indexes that account only for down-side risk (i.e. 
movements of farm income on the left side of the distribution) (Horcher, 2005) or com-
mon variability indexes considering both sides of the distribution. On the other hand, the 
paper assesses how important it is to account for the heterogeneity existing within the 
farm sector. This issue is relevant because of two main reasons. First, as it is well known, 
focusing on regional or national aggregates is expected to generate aggregation bias that 
underestimates the level of variability experienced by farmers (Coble et al., 2007). This 
aspect is relevant also to correctly assess the extent of the compensations of income loss-
es that the recently introduced CAP income stabilization tool will pay to farmers. This is 
especially because it is still not very clear how the reference income and deviations from it 
will be assessed provided the existing heterogeneity in data availability and quality among 
EU Member States. Second, because farms strongly differ in terms of several dimensions, 
different farm groups can face different levels, types and evolution of variability. 

The paper contributes to the existing literature because, based on our current knowl-
edge, empirical evidences based on individual farm income data on the use of down-side 
risk indicators, on the extent of the aggregation bias and on whether income variability 
has increased over time are scant  in the agricultural economics literature. Furthermore, 
the analysis is innovative because, to test for differences in terms of income variability 
between farm groups and periods, it uses non-parametric approaches that are less affected 
than parametric tests by the presence of outliers – a situation that is often encountered 
when using individual farm data.

The results of the analysis could also feed the policy debate regarding whether the 
instability of farm income is a relevant policy concern and the introduction of the new 
CAP risk management measures is justified. The analysis also provides insights on how to 
target intervention on the farm groups where it is more needed.

The next section casts the analysis into the previous literature on farm income vari-
ability while the following one describes data and methodology. Section 4 presents the 

1 Regulation (EU) No 1305/2013 of the European Parliament and of the Council of 17 December 2013 (O.J.E.U. 
L 347 of the 20.12.2013) establishes risk management measures that cover: (a) financial contributions to premi-
ums for farm insurances (Art. 37); (b) financial contributions to mutual funds (Art. 38); (c) an income stabilisa-
tion tool, in the form of financial contributions to mutual funds, providing compensation to farmers for a severe 
drop in their income (Art. 39). 



65The instability of farm income

obtained empirical results while the last paragraph critically discusses the main findings 
of the analysis, underlines its weaknesses and identifies possible areas for further research.

2. Literature review on farm income instability

Several risks affect farm businesses but most of the analyses focus on the business risk 
that is generated by the aggregate effect of production, market and other sources of busi-
ness specific risks (Hardaker et al., 2007). While many analyses have been focused on sin-
gle farm risk sources (e.g. yield risk or price risk) or single production enterprises (e.g. 
milk production), what is relevant is the interaction among the many elements generating 
business risk (OECD, 2009). This is because the overall risk a farmer is facing depends 
on the interactions among the different production activities carried on-farm and on the 
evolution of different parameters (e.g. product prices and yields). Furthermore, in the EU, 
farm incomes are supported by mean of direct payments that represent around 30% of 
farm income (European Commission, 2011) and have been claimed to reduce income var-
iability (Agrosynergie, 2011; Cafiero et al., 2007; El Benni et al., 2012). 

These elements support the idea that, to evaluate the business risk a farmer is fac-
ing, it is needed to account for the variability of the overall income of his/her farm over 
time (Mishra and Sandretto, 2002; OECD, 2009). In almost all the agricultural economic 
literature, analyses on farm risk refer to income (Agrosynergie, 2011; El Benni et al., 2012; 
El Benni and Finger, 2013; European Commission, 2011; Finger and El Benni, 2014; Meu-
wissen et al., 2008; OECD, 2003, 2009; Vrolijk and Poppe, 2008; Vrolijk et al., 2009). How-
ever, the variability of the economic performances of firms can also be analysed by using 
cash flow indicators2 (Plewa and Friedlob, 1995). Some Authors have supported the idea 
that cash flow can be used to do so because of two main reasons. First, cash flow is bet-
ter observable and harder to manipulate under generally accepted accounting principles. 
Second, it is closer to liquidity management and can be a good indicator for the analysis 
of the firm’s survival: a company that is short on cash could fail and be technically bank-
rupt despite it has a large amount of accounts receivable on its balance sheet. Despite this, 
cash flow has not been used yet in the agricultural economic literature apart in few cases 
(Meier, 2004). Because of this, we have decided to focus on farm income to allow for the 
comparability of the results with those of previous analyses. 

In order to assess the level of instability farmers are facing, it is preferable to have 
farm-level time-series because the aggregation of data from different farms generates 
aggregation bias. At higher levels of aggregation, poor incomes in some farms are offset 
by good incomes in others thereby reducing the overall variability (Coble and Dismukes, 
2008; Finger, 2012; OECD, 2009)3. Several Authors conclude that using aggregated data 
can severely underestimate the farm level risk (Coble et al., 2007; Coble and Dismukes, 
2008; Kimura et al., 2010; Popp et al., 2005). Despite this,  empirical evidences on the 
extent of aggregation bias in the case of farm income variability are scant.

2 We thank one of the anonymous reviewers for suggesting us to consider this branch of literature. The use of 
cash flow seems a very promising and innovative direction for future research developments. In particular, it 
could be very interesting to compare income variability with cash flow variability. 
3 However, this phenomenon is reduced when systemic risk is pervasive and relevant.



66 S. Severini, A. Tantari, G. Di Tommaso

This paper focuses on business related risks and considers only farm income. This 
seems coherent with the sectorial nature of CAP. However, other analyses, such as Mishra 
and Sandretto (2002), have investigated the instability of the income of farm households, 
i.e. considering also off-farm income. Mishra and Sandretto (2002) showed that the vari-
ability of the incomes of farm families has not diminished over the considered 7 decades. 
However, their analysis relies on national-wide data and does not provide evidences about 
differences within the sector4. Farm level analyses in the US focus more on the decompo-
sition of household income variability by income sources than on the level of income vari-
ability per se (Mishra and El-Osta, 2001; Mishra et al., 2002). 

Empirical evidences based on single farm data are not abundant also in Europe. The 
analyses by Vrolijk and Poppe (2008) and Vrolijk et al. (2009) represent relevant pieces 
of literature on this issue. These rely on large samples of farms, have been developed in 
a considerable number of EU countries and allow for comparison between countries and 
types of farming. However, differences between countries and types of farms have not 
been subject to statistical testing. Finally, the focus in the EU is in most of the cases on 
farm business income (i.e. off-farm incomes are not considered) because of data availabil-
ity constraints and the agricultural policy orientation of the analyses. However, this is not 
the case of recent analyses developed in Switzerland where the national farm data network 
also collects data on off-farm incomes (El Benni et al., 2012; Finger and El Benni, 2014).

In order to assess and to manage risk, it has been found that, in some cases, it can be 
preferable to consider down-side risk other than common variability indexes (Miller and 
Leiblein, 1997). This is because farmers are generally more concerned with movements of 
farm income on the left side of the distribution (Horcher, 2005). However, indexes con-
sidering both sides of the distribution could perform equally well whenever, for example, 
the distribution of income over time is symmetric. Thus, the use of one type of variability 
index or the other should be chosen on the basis of the specific situation under study.

3. Data and methodology

3.1 Data

Having to assess the variability of farm economic results over the years, there is the 
need to select farms that have been in the samples for a reasonably long period of years. 
The analysis is based on data from all individual farms that belonged to the whole Italian 
sample of the Farm Accountancy Data Network (FADN) during all years of the decade 
2003-2012. Thus, the dataset is made by a balanced panel because the considered farms do 
not change over the 10 years. These are 2404 farms for 24040 observations. 

Referring to a constant sample of farms and the same time interval allows for better 
comparing results among farm groups and sub-periods because this avoids that some of 
the reported differences may be due to changes in farm composition5. The resulting num-

4 Mishra and Sandretto (2002) use individual farm data in the second part of their paper to verify that off-farm 
income has contributed to the farm household income stability.
5 The use of an unbalanced panel dataset, while increasing the number of observations, could generate compara-
bility problems. This is because farms refer to time intervals of different length and to different periods (e.g. at 
the beginning or at the end of the interval of time).



67The instability of farm income

ber of farms is large enough also for comparing groups of farms selected within the sam-
ple. This is important because the considered farms have been grouped according to types 
of farming, economic size and productivity level (European Commission, 2010) (Table 1). 
7 types of farming (TF) have been considered to account for production orientation and 
specialisation. Economic size refers to small, medium and large farms defined by mean of 
the European Size Unit (ESU) classes provided by FADN. Finally, farms have been also 
classified into four groups according to the level of a partial productivity index calculated 
as the ratio between farm income and the amount of labour used on farm in terms of 
Annual Work Units (AWU) (European Commission, 2010).

Unfortunately, the choice to have only farms belonging to all considered 10 years has 
driven us to have a not randomly selected sub-sample. This has two important consequenc-
es. On the one hand, the selected sub-sample cannot be considered representative of the 
whole farm population. On the other hand, the statistical weights provided by FADN annu-
ally for each sampled farm cannot be used for reporting the results to the farm population. 
However, despite these limitations, it is important to note that the distribution of the farms 
within the sub-sample is very similar to the distribution of farms within the whole sample 
when grouped by types of farming, macro-regions and altimetry zones (Table A1 and A2 
in the Appendix). The Finger and Kreinin (1979) similarity indexes computed on the two 
samples show a level of similarity that is never below 90%6. This suggests that the sub-sam-
ple does not provide an incomplete representation of the Italian farming sector. 

3.2 Income definition and treatment of trends

The focus is on the FADN variable Farm Net Income (FI) that is the remuneration 
to fixed factors of production of the family (work, land and capital) and remuneration 
to the entrepreneur’s risks (loss/profit) in the accounting year (European Commission, 
2007). FADN is a business oriented database, thus it provides data on the income com-
ing from the farming activities but it does not provide data on off-farm income earned 
by farm family members. However, it includes returns from nonfarm-based production 
activities such as, for example: hiring out of equipment, agro-tourism and forestry activi-
ties. FI is net of taxes linked to the farming activity but does not deduct personal taxes 
that are very much dependent on the overall amount of income (both on and off-farm) 
of the family members. The size of the FI depends, among others, on the relative amount 
of factors owned by the family provided that FI is net of the wage, rent and interest paid 
to third parties. Thus, if a farmer decides to use a large amount of external factors, this 
implies that (ceteris paribus) the remaining FI declines and, in some cases, it is likely that 
it becomes also more variable7.

6 The Finger – Kreinin index has been originally developed to compare the structure of the export of products of 
two countries. It sums the shares of all products considering, for each product, the minimum value between the 
two series. Thus, it assumes a value of 100% in the case of complete similarity, while it tends to zero as long as 
similarity declines.
7 The choice to use external factors is a management decision and farms indeed differ because they use different 
management strategies. The choice of obtaining labour, land and capital from third parties affects farm economic 
performances, their variability over time and, in turns, the wellbeing of farm families. Thus, it seems logical that 
different management strategies have different implications also in terms of the income risk farmers face.



68 S. Severini, A. Tantari, G. Di Tommaso

As shown in the previous paragraph, farm income has been the economic variable 
used to assess the farm risk by almost all the analyses developed on the EU farm sector. 
This is because farm income describes better than other variables (e.g. revenues) the eco-
nomic performances of a specific farm provided that, at the end, farmers are interested in 
how much the resources they use on-farm are remunerated and how this remuneration 
varies over time.

The fact that the mean or expected value of farm income has a trend or a cyclical 
behaviour does not necessarily imply risk: an economic variable may follow well-defined 
patterns that are known to farmers (OECD, 2009). Trends in income may occur because, 
for example, prices generally increase over time due to inflation and trends are pervasive 
in crop yields. For this reason, it has been chosen to eliminate the impact of inflation and 
to assess the variability around the trend (if existing). The original FI series have been first 
deflated by using the GDP deflator and later standardised (dividing each value by the 10 
year average) to have all series centred around 1. The series have been then explored to 
identify linear trends by pooling all farms into 7 Types of Farming (TF) (i.e. farms with 
a similar production pattern) (European Commission, 2010) (Table 1). The trends have 
been estimated by using a robust regression approach based on two weight functions 
(Huber weights and bi-weights) to account for the presence of outliers (Finger and Hedi-
ger, 2008; Huber, 1964; Maronna et al., 2006). Because in all types of farming but special-
ised granivore farms significant linear trends have been identified, the deflated FI series 
have been detrended in those 6 cases8. 

3.3 Variability indexes

The detrended series have been used to calculate two variability indexes in each farm. 
These are Standard Deviation (V1) and Semi-Standard Deviation (V2) of farm income 
over the decade9. For a generic i-th farm, this latter index has the following structure:

V 2i =
tFIi ,t<FI i

T

∑
 FIi,t − FI i( )

2

m
 (1)

8 Because we worked with short time-series, it was considered not possible to get sound results from models able 
to explore the possible cyclical nature of the series. However, the use of linear trend appears still valid provided 
that estimation results are satisfactory. Furthermore, while the cyclical nature of price is very important, this is 
not necessarily the case for income series. This is because farm revenues come from different activities (i.e. diver-
sified farms), they are also influenced by the evolution of unitary production levels (e.g. yields) and, in some 
cases, there is a natural hedge between the two (i.e. negative correlation between price and yield). Furthermore, 
incomes are also influenced by the evolution of production costs. In the considered decade, the prices of some 
commodities have shown some cyclical path (e.g. cereal prices higher than usual for two or three years) (Figure 
A1 in the Appendix). However, the examination of the evolution of farm income by types of farming does not 
provide evidences of clear cyclical behaviour (Table A3 in the Appendix). This is also the case of field crop farms 
because, in the same period in which cereal prices have been high, costs have also reached high levels due to the 
extraordinary high prices of some required inputs, noticeably, chemical fertilisers (Figure A2 in the Appendix).
9 Standard Deviation has been calculated on standardized data, so it can be directly interpreted as a Coefficient 
of Variation of the absolute income. This allows for the direct comparability between farm groups and periods.



69The instability of farm income

It differs from V1 because it considers only data from those years (m < T) in which 
income levels are below the average value FIi,t < FI i  that, in our case, is calculated over 
the 10 year observations (T=10). Thus, this index focuses only on the down-side risk (i.e. 
adverse income conditions). 

However, V1 could be a good proxy of V2 and has a very desirable property: it relies 
on a larger number of observations than V2. This latter aspect seems relevant in our case 
because only 10 years of data are available.

Variability indicators V1 and V2 have been obtained for each farm of the sample. The 
mean and median levels of variability have been analysed by grouping farms according to 
the three dimensions previously described: type of farming; economic size; farm produc-
tivity (Table 1). The disaggregation by types of farming is interesting because some pro-
duction industries are inherently more risky than others and because diversification can 
reduce risk. Furthermore, the relative level of support provided by the CAP direct pay-
ments strongly differs in Italy among farms with different production patterns and eco-
nomic size, and this source of income has been found to stabilise farm income (Cafiero 
et al., 2007; El Benni et al., 2012; Enjolras et al., 2014; OECD, 2009; Severini and Tantari, 
2013). Referring to productivity level is also relevant because the same relative level of 
income variability can have more severe implications for low productivity farms than for 
high productivity farms. This is because owned resources already receive a low remunera-
tion and this generally implies a limited capability to accumulate cash reserves to cope 
with low income years.

The distributions of V1 and V2 within the groups and within the whole sample have 
been explored calculating common dispersion indexes (Standard Deviation and Coeffi-
cient of Variation) and testing for normality by means of the  Shapiro-Francia test (Shap-
iro and Wilk, 1965). The results of these tests allow to reject the normality of the distribu-
tions. This has suggested to focus on median values (other than on mean values) of each 
considered group and to rely on non-parametric tests for comparisons among variability 
indexes, groups and periods.

Correlations between V1 and V2 have been calculated for the whole sample and for 
each group using the Spearman’s test and the related rho correlation coefficient (Hauke 
and Kossowski, 2011). 

Differences between farm groups have been tested by means of both Kruskal-Wallis 
test and Wilcoxon rank-sum test (Kruskal and Wallis, 1952; Mann and Whitney, 1947). 
The first allows to test for the presence of at least one inequality among groups. When the 
results of this test allow to reject the null hypothesis, the pairwise comparisons between 
groups has been developed by using Wilcoxon rank-sum tests. This latter has been also 
used to test for differences between variability indexes calculated in the periods 2003-2007 
and 2008-2012.

3.4 Aggregation bias

In order to have a reference level to assess aggregation bias, V1 has been calculated 
not just on individual farm data but also on the average income of different groups of 
farms. The average income of the j-th group of farms (e.g. a specific type of farming) in 
the t-th year is:



70 S. Severini, A. Tantari, G. Di Tommaso

AFI j ,t = i j

n j
∑ FIi,t

nj
 (2)

where nj is the number of farms belonging to the j-th group. The average over time of 
such aggregated income indicator is:

AFI j = 1
10 t=1

10

∑AFI j ,t  (3)

Thus, when using the average income of the j-th group of farms10, V1 is calculated as 
it follows:

V1j =
t=1

10

∑
AFI j ,t − AFI j( )

2

10
 (4)

The levels of variability calculated as in (4) are referred to as Aggregated data in Table 
2. The other columns of Table 2 refer to basic statistics of the variability figures calculated 
on each individual farm of the sample.

4. Empirical results

While the median levels of standard deviation (V1) are higher than those of the semi-
standard deviation (V2) (Table 1), the level of the two variability indicators are very much 
correlated. In all groups it is possible to reject the null hypothesis that V1 and V2 are not 
correlated according to the Spearman’s test at 1% significance level. The rho correlation 
coefficient calculated on the whole sample is 0.889 and is greater than 0.79 in all groups 
(Table 1). These results allow to conclude that, the two indicators provide a very similar 
representation of the relative income risk of the considered individual farms. This  sup-
ports the hypothesis that there is not an apparent advantage to rely on a down-side risk 
indicator (V2) in similar circumstances. Because of this, the presentation of the empiri-
cal results will focus on V1 only even because, as previously noted, V1 is calculated on a 
larger number of observations than V2.

The median of V1 for the total sample (0.637) suggests that farmers are faced with 
relatively high levels of income variability (Table 1). However, there is a high dispersion 
of V1 levels within each group as testified by the large values of standard deviation (SD) 
and coefficient of variation (CV) of V1 (Table 2). This is particularly the case of specialist 
field crop farms and, to a lesser extent, of specialist grazing livestock farms and specialist 
granivore farms (Table 2). 

10 V1 has been also calculated on the average income of all farms of the sample (i.e. not just in each specific 
group of farms) by using the same approach. This figure is reported in row Total sample of column Aggregated 
data  of Table 2. 



71The instability of farm income

The comparison of the V1 calculated on individual farm data with those calculated 
on aggregated data allows to assess the extent of the potential aggregation bias. V1 cal-
culated on aggregated data are always way lower than the medians of the values of V1 
calculated on individual farm data. On the whole sample, the variability calculated on the 
aggregated data is less than ¼ of the median value of farm level variability (Table 2). Simi-
lar results are found even when the sample is subdivided according to types of farming: 
aggregation bias seems relatively lower for specialist field crop and for specialist granivore 
farms. However, in all cases the medians V1 calculated on aggregated data do not exceed 
3/5 of those calculated on individual farm data (Table 2). This means that the extent of the 
aggregation bias is not substantially reduced even when farms are grouped according to 
their production specialisation.

Despite the large dispersion of variability within each farm group, it is possible to say 
that groups differ. Variability is lower in specialist grazing livestock farms (TF 4) and spe-
cialist horticulture farms (TF 2), while higher in specialist granivore farms (TF 5) (Table 2).

Table 1. Sample size, median values and correlation between Standard Deviation (V1) and Semi-
Standard Deviation (V2) of farm income within each group and the whole sample of farms.

Sample 
size

Standard 
Deviation

Semi 
Standard 
Deviation

Number V1 V2 Spearman's 
rho^

Types of Farming (TF) Code
Specialist field crops 1 572 0.659 0.534 0.911 ***
Specialist horticulture 2 276 0.584 0.507 0.790 ***
Specialist permanent crops 3 715 0.665 0.550 0.895 ***
Specialist grazing livestock 4 493 0.572 0.484 0.887 ***
Specialist granivore 5 84 0.844 0.739 0.796 ***
Mixed cropping 6 161 0.688 0.564 0.879 ***
Mixed livestock and Mixed 
crops-livestock

7 103 0.658 0.567 0.904 ***

Economic size (ESU classes)
Small (Classes 1, 2 and 3) 699 0.715 0.595 0.895 ***
Medium (Classes 4, 5 and 6) 1595 0.604 0.509 0.887 ***
Large (Classes 7 and 8) 110 0.686 0.589 0.828 ***

Productivity levels (FI per unit of labour)
Low 601 0.857 0.708 0.873 ***
Low-Medium 601 0.642 0.530 0.880 ***
Medium-High 601 0.549 0.479 0.880 ***
High 601 0.545 0.457 0.867 ***

Total sample 2404 0.637 0.530 0.889 ***

Correlation 
between V1 and 

V2:

^ In all groups it is possible to reject the null hypothesis that V1 and V2 are not correlated according 
to the Spearman’s test at 1% significance level (***).
Source: Own elaborations on a constant sample of the whole Italian FADN farms for the period 2003-
2012.



72 S. Severini, A. Tantari, G. Di Tommaso

According to the Kruskal-Wallis test, there is at least one inequality among types of 
farming. The results of Wilcoxon rank sum test show that specialist granivore farms (TF 5) 
face a level of variability that is significantly higher than in the other groups. Specialist hor-
ticulture farms and specialist grazing livestock farms are in the opposite situation (Table 3).

The high variability found in specialised granivore farms is consistent with the find-
ings of previous studies (Vrolijk and Poppe, 2008; Vrolijk et al., 2009; European Commis-
sion, 2011) and can be explained by the nature of these farms: high specialisation and a 
limited importance of CAP direct payments. Similarly, the low income variability of spe-
cialist grazing farms is consistent with the findings of Vrolijk and Poppe (2008). On the 
contrary, the finding that also specialist horticulture farms are facing a relatively low level 
of income variability is not fully consistent with previous studies and contrasts with the 
common believe that horticulture crops are affected by high business risk. In this case, 
it is possible that diversification and other risk management tools already used by such 
farms (e.g. crop insurance) are effective tools to stabilise income.

Differences exist also among farms of different size. The median V1 in small farms is 
greater than in large farms and almost double than in medium farms (Table 4). This result 
is coherent with what has been found by the European Commission (2011). Despite the 
dispersion of values within the groups, Kruskal-Wallis tests suggest that there is at least 
one inequality among these groups. Furthermore, the Wilcoxon rank sum tests show that 
income variability of medium size farms is lower than that of the other two groups of 
farms with a statistical significance of 1%. This means that small farms are facing a rela-
tively higher level of income variability. As suggested by Vrolijk and Poppe (2008), this 
could be caused by the fact that income of small farms is in many cases very low, so that 

Table 2. Income variability levels (V1) by type of farming and in the whole sample. V1 has been calcu-
lated on aggregated data (i.e. sum of individual farm data) and on individual farm data.

Aggregated 
data^

V1 Mean Median SD^^ CV^^

Description Code
Specialist field crops 1 0.400 1.297 0.659 8.535 6.58
Specialist horticulture 2 0.091 0.689 0.584 0.479 0.69
Specialist permanent crops 3 0.120 0.909 0.665 1.523 1.67
Specialist grazing livestock 4 0.099 0.787 0.572 1.780 2.26
Specialist granivore 5 0.469 1.343 0.844 3.426 2.55
Mixed cropping 6 0.153 0.982 0.688 1.320 1.34
Mixed livestock and Mixed 
crops-livestock

7 0.233 1.111 0.658 1.699 1.53

Total sample 0.150 0.980 0.637 4.401 4.49

Types of Farming (TF):

of V1

Individual farm data

^ As specified in (4), figures in the column Aggregated data refer to the V1s calculated on the averag-
es of the incomes of the farms belonging to each Type of farming as well as of all farms in the sample 
(Total sample). 
^^ SD and CV of V1 should be considered with caution because V1 is not-normally distributed.
Source: Own elaborations on a constant sample of the whole Italian FADN farms for the period 2003-
2012.



73The instability of farm income

small changes in revenue (or costs) can cause high relative changes in income. Howev-
er, despite that the median value is higher in small than in large farms, the results of the 
Wilcoxon test do not support the hypothesis that there are significant differences between 
these two groups of farms while it suggests that income is more stable in medium size 
farms than in the other two groups (Table 4)11.

Income variability levels also differ among farms with different levels of productiv-
ity (i.e. farm income per unit of labour). The highest variability can be found in the low 
productivity farms (Table 5). Similarly, income variability in farms belonging to the next 

11 In very large farms most of the labour is provided by non-family members who receive salaries accounted for 
as explicit costs. This makes income of those farms relatively small if compared with the total revenues generated. 
Thus,  also in this case, small changes in revenue (or costs) can cause high relative changes in income. The oppo-
site can be said for medium size farms where a large share of the labour is often provided by family members.

Table 3. Significance of the Wilcoxon rank sum test between couples of types of farming.

Types of Farming (TF) Code 1 2 3 4 5 6 7
Specialist field crops 1 *** *** ***
Specialist horticulture 2 *** *** *** *
Specialist permanent crops 3 *** ***
Specialist grazing livestock 4 *** *** **
Specialist granivore 5 *** ***
Mixed cropping 6
Mixed livestock and Mixed 
crops-livestock

7

Types of farming codes:

^ Significant at 1% (***), 5% (**) and 10% (*) according to Wilcoxon rank sum tests.
Source: Own elaborations on a constant sample of the whole Italian FADN farms for the period 2003-
2012.

Table 4. Income variability levels by economic size classes and in the whole sample. Mean, Median, 
Standard Deviation and Coefficient of Variation of V1.

Mean  Median SD^ CV^

Medium Large

Small 1, 2 and 3 699 1.404 0.715 7.90 5.63 ***
Medium 4, 5 and 6 1595 0.789 0.604 1.06 1.35 **
Large 7 and 8 110 1.049 0.686 2.94 2.80
Total sample 2404 0.980 0.637 4.40 4.49

of V1

Wilcoxon rank sum 
test results^^:Sample 

size
ESU 

Classes

^ SD and CV values should be considered with caution because V1 is not-normally distributed. 
^^ Significant at 1% (***), 5% (**) and 10% (*).
Source: Own elaborations on a constant sample of the whole Italian FADN farms for the period 2003-
2012.



74 S. Severini, A. Tantari, G. Di Tommaso

group (i.e. low-medium productivity) is significantly higher than in those belonging to the 
other half of the distribution (Table 5). The fact that the income is less stable in farms 
already facing productivity problems suggests that these farms are even more endangered 
by income variability than other farms. This is because a low level of productivity often 
results in the under-remuneration of farm resources and does not allow to accumulate 
cash reserves to be used in low income years.

Variability has increased over time. The medians of V1 calculated over the period 
2008-2012 in the whole sample and in each farm group are always greater than those cal-
culated in the previous period (2003-2007) (Table 6). This result is also supported by the 
analysis of the mean and the median values of V1 calculated over 6 overlapping sub-peri-
ods (Table 7). In both cases, there is a slight increase of the values of V1 moving from the 
first to the following periods, except for the last one.

The largest increment is found in specialist horticulture farms (TF 2). Accord-
ing with the results of the Wilcoxon rank sum tests, there are statistical significant dif-
ferences between the two periods in all TF but specialist permanent crops farms (TF 3) 
that showed the lowest increase (Table 6). Significant increases can be found also in the 
three size groups. Large farms suffered the greatest increase, while small farms the low-
est (Table 6). However, it is important to note that income variability increased the most 
in the farms with the highest levels of productivity while the less in those with the lowest 
levels of productivity (Table 6). This shows an improvement of the relative condition of 
low productivity farms.

5. Conclusions

The paper has provided empirical evidences regarding the five research issues 
described in the introduction. First, in the considered case, there is not an apparent and 
relevant advantage to account only for down-side risk to assess the relative level of income 
risk faced by farmers. This supports the hypothesis that simple variability indexes, such 
as SD and CV of farm income, could be used under similar circumstances. Second, a 

Table 5. Income variability levels by classes of productivity levels (FI per unit of labour) and in the 
whole sample. Median and Coefficient of Variation (CV) of V1.

 Median CV

25-50% 50-75% 75-100%

Low 0 - 25% 0.857 5.01 *** *** ***
Low-Medium 25-50% 0.642 1.34 *** ***
Medium-High 50-75% 0.549 1.86
High 75-100% 0.545 0.70
Total sample 0.637 4.49

Productivity classes
Wilcoxon rank sum test 

results:

of V1

^Significant at 1% (***), 5% (**) and 10% (*) according to Wilcoxon rank sum tests .
Source: Own elaborations on a constant sample of the whole Italian FADN farms for the period 2003-
2012.



75The instability of farm income

large potential aggregation bias arises when using aggregated data. This confirms that it is 
preferable to use individual farm data to avoid underestimating the income risk farmers 
are facing. Third, Italian farms seem affected by a not negligible level of income variabil-
ity and this is the case even if variability is calculated on deflated and de-trended series. 
Fourth, significant differences among farm groups have been found in terms of levels 
and evolution of income variability. This implies that not all farms face the same level of 
income risk and urge policies supporting their participation to risk management meas-
ures. In particular, income variability seems relatively lower in specialist horticulture and 

Table 6. Variability levels in the periods 2003-2007 and 2008-2012 for the whole sample and for each 
farm group. Means and Medians of V1 and relative change over time of the Medians.

Change in  
median 
values

Mean Median Mean Median %
Code

Specialist field crops 1 1.270 0.515 0.948 0.582 *** 13.0%
Specialist horticulture 2 0.424 0.302 0.719 0.565 *** 87.2%
Specialist permanent crops 3 0.700 0.503 0.833 0.513 2.0%
Specialist grazing livestock 4 0.574 0.378 0.730 0.457 *** 21.1%
Specialist granivore 5 0.817 0.515 1.450 0.648 ** 25.7%
Mixed cropping 6 0.664 0.448 0.990 0.588 *** 31.2%
Mixed livestock and Mixed 
crops-livestock

7
0.857 0.429 0.954 0.593 *** 38.1%

Farm size classes (ESU classes)
Small (Classes 1, 2 and 3) 1.198 0.531 1.160 0.585 *** 10.1%
Medium (Classes 4, 5 and 
6) 0.593 0.424 0.740 0.509

***
20.1%

Large (Classes 7 and 8) 0.983 0.438 0.768 0.549 ** 25.4%
Productivity classes (FI per unit of labour)

Low 1.454 0.665 1.422 0.701 *** 5.4%
Low-Medium 0.649 0.465 0.802 0.524 ** 12.5%
Medium-High 0.590 0.380 0.602 0.438 *** 15.1%
High 0.452 0.354 0.627 0.485 *** 36.8%

Total sample 0.786 0.449 0.863 0.531 *** 18.1%

2003-2007 2008-2012

Types of Farming (TF)

^ Significant at 1% (***), 5% (**) and 10% (*) according to Wilcoxon rank sum tests.
Source: Own elaborations on a constant sample of the whole Italian FADN farms for the period 2003- 
2012.

Table 7. Variability levels of income over the 6 sub-periods. Mean and median of V1 calculated on the 
whole sample. 

2003- 
2007

2004- 
2008

2005- 
2009

2006- 
2010

2007- 
2011

2008- 
2012

Mean 0.786 0.786 0.830 0.857 0.892 0.863
Median 0.449 0.505 0.539 0.559 0.571 0.531

Source: Own elaborations on a constant sample of the whole Italian FADN farms for the period 2003-
2012.



76 S. Severini, A. Tantari, G. Di Tommaso

grazing livestock farms than in other types of farming. On the contrary, the analysis has 
confirmed that income variability is higher in small than in large farms. Farms with rela-
tively lower productivity levels are affected by a relatively larger income variability. Fifth, 
results suggest that income variability has increased over the considered decade.

All these results have policy implications. The relevant aggregation bias and the dif-
ferences within farms groups show that the recently introduced Income Stabilisation Tool 
correctly refers to the income of individual farms (i.e. does not rely on more aggregat-
ed data) for assessing income evolutions and the indemnities to be paid. The high and 
increasing level of income variability supports the idea that the introduction of risk man-
agement tools within the CAP toolbox could be justified. However, it is worth noting that 
income risk can be also coped by, for example, accumulating enough cash reserves or by 
integrating farm income with off-farm income. The differences among farm groups sug-
gest that there is scope for targeting income stabilising policies on those farm types where 
income is more unstable and that are less able to manage income risk. Empirical results 
suggest that low productivity farms are likely those with the most pressing needs. These 
have been found to have a relatively high level of income variability and are very likely 
more vulnerable than farms with higher productivity levels even because are less able to 
accumulate large cash reserves (Mishra et al., 2002).

Unfortunately, the analysis suffers from at least three main limitations that should be 
overcome in future analyses. First, the selected sub-sample cannot be considered representa-
tive of the whole farm population because it is not randomly selected. Furthermore, the 
results cannot be reported to the farm population based on representative weights. Therefore, 
sample statistics can differ from the population parameters. However, because the distribu-
tion of the farms within the sub-sample is very similar to the distribution of farms within 
the whole sample in terms of several dimensions, it is possible to say that the considered sub-
sample does not provide an incomplete representation of the Italian farming sector. The sec-
ond limitation is that, as previous studies on the same topic, it does not investigate the causes 
of the assessed income variability. Thus it seems important to develop future research to dis-
entangle the role of different income components including CAP direct payments. 

The third limitation is that, as in most of the analysis developed so far in the EU, the 
analysis accounts only for farm business income and this arises three main problems. 
First, a high variability of farm income does not necessarily imply an high farm household 
income variability provided that off-farm incomes have been found to stabilize the income 
of farm households (Mishra and Sandretto, 2002). Second, farmers can manage income 
instability by accumulating cash reserves to be used in low income years. Third, cash flow 
variability can be a different but promising way to look at the risk farmers are facing. 
These last three issues suggest that future analyses should enlarge the focus beyond what 
has been done in this paper by considering off-farm incomes (when data will become 
available), the availability and use of cash reserves and by comparing the variability of 
income with that of cash flow indicators.

References

Agrosynergie (2011). Evaluation of income effects of direct support. Final report. Brux-
elles, May.



77The instability of farm income

Cafiero, C., Capitanio, F., Cioffi, A. and Coppola, A. (2007). Risk and Crises Management 
in the reformed European Agricultural Crises Policy. Canadian Journal of Agricul-
tural Economics, Special Issue December 2007: 419-441.

Coble, K.H., Dismukes, R. and Thomas, S. (2007). Policy Implications of Crop Yield and 
Revenue Variability at Differing Levels of Disaggregation. Paper presented at the 
American Agricultural Economics Association Annual Meeting, Portland, Oregon, 
July 29-August 1, 2007.

Coble, K.H. and Dismukes, R. (2008). Distributional and Risk Reduction Effects of Com-
modity Revenue Program Design. Applied Economic Perspectives and Policy 30(3): 
543-553.

El Benni, N., Finger, R. and Mann, S. (2012). Effects of agricultural policy reforms and 
farm characteristics on income risks in Swiss agriculture. Agricultural Finance 
Review 72(3): 301-324.

El Benni, N. and Finger, R. (2013). Gross revenue risk in Swiss dairy farming. Journal of 
Dairy Science 96: 936-948.

Enjolras, G., Capitanio, F., Aubert, M. and Adinolfi, F. (2014). Direct payments, crop 
insurance and the volatility of farm income. Some evidence in France and Italy. New 
Medit 1(2014): 31-40.

European Commission (2007). Farm Accountancy Data Network. Definitions of variables 
used in FADN standard results. Available at: http://ec.europa.eu/agriculture/rica.

European Commission (2010). Farm Accountancy Data Network. An A to Z of methodol-
ogy. Available at: http://ec.europa.eu/agriculture/rica/pdf/site_en.pdf.

European Commission (2011). Farm Economics brief N°1. Income developments in EU 
farms. Bruxelles, June 2011.

Finger, R. and Hediger, W. (2008). The application of robust regression to a production 
function comparison. The Open Agriculture Journal 2: 90-98.

Finger, R. (2012). Effects of crop acreage and aggregation level on price‐yield correlations.  
Agricultural Finance Review 72(3): 436-455.

Finger, R. and El Benni, N. (2014). A Note on the Effects of the Income Stabilisation Tool 
on Income Inequality in Agriculture. Journal of Agricultural Economics 65: 739-745.

Finger, J.M. and Kreinin, M.E. (1979). A Measure of `Export Similarity’ and Its Possible 
Uses. The Economic Journal 89(356): 905-912.

Hardaker, J.B., Huirne, R.B.M., Anderson, J.R. and Lien, G. (2007). Coping with risk in 
agriculture. Second edition. CABI Publishing, Wallingford (UK).

Hauke, J. and Kossowski, T. (2011). Comparison of values of Pearson’s and Spearman’s 
correlation coefficients on the same sets of data. Questiones Geographicae 30(2) 
87-93.

Horcher, K.A. (2005). Essentials of financial risk management. John Wiley & Sons.
Huber, P.J. (1964). Robust estimation of a location parameter. Annals of Mathematical Sta-

tistics 35: 73-101.
Kimura, S., Antón, J. and LeThi, C. (2010). Farm Level Analysis of Risk and Risk Manage-

ment Strategies and Policies: Cross Country Analysis. OECD Food, Agriculture and 
Fisheries Working Papers, No. 26, OECD Publishing.

Kruskal, W.H. and Wallis, W.A. (1952). Use of ranks in one-criterion variance analysis. 
Journal of the American Statistical Association 47 (260): 583-621.



78 S. Severini, A. Tantari, G. Di Tommaso

Mann, H.B. and Whitney, D.R. (1947). On a Test of Whether one of Two Random Variables 
is Stochastically Larger than the Other. Annals of Mathematical Statistics 18(1): 50-60.

Maronna, R.A., Martin, R.D. and Yohai, V.J. (2006). Robust statistics. Theory and meth-
ods. New York, Wileys and Sons.

Matthews, A. (2010). Perspectives on Addressing Market Instability and Income Risk for 
Farmers. IIIS Discussion Paper, 342, Dublin, April.

Meier, B. (2004). The role of cash flow indicators in understanding farm households. 
Paper presented at the OECD-workshop, 29/30 April 2004, Paris (France). Avail-
able at: http://www.oecd.org/tad/agricultural-policies/32022387.pdf. Accessed on: 
November 30th, 2015.

Meuwissen, M., van Asseldonk, M., Pietola, K.S., Hardaker, J.B. and Huirne R. (2011). 
Income Insurance as a Risk Management Tool After 2013 CAP Reforms? Europe-
an Association of Agricultural Economists Congress, Zurich, Switzerland, August 
30-September 2.

Meuwissen, M.P.M., van Asseldonk, M.A.P.M. and Huirne, R.B.M. (eds) (2008). Income 
stabilisation in European agriculture. Wageningen Academic Publishers, Wagenin-
gen, The Netherlands.

Mishra, A.K. and El-Osta, H.S. (2001). A temporal comparison of sources of variability in 
farm household income. Agricultural Financial Review 61(2): 181-198. 

Mishra, A.K., El-Osta, H.S., Morehart, M.J., Johnson, J.D. and Hopkins, J.W. (2002). 
Income, Wealth, and the Economic Well-Being of Farm Households. Farm Sec-
tor Performance and Well-Being Branch, Resource Economics Division, Economic 
Research Service, U.S. Department of Agriculture. Agricultural Economic Report 
No. 812.

Mishra, A.K. and Holthausen, D.M. (2002). Effect of farm income and off-farm wage vari-
ability on off-farm labor supply. Agricultural and Resource Economics Review 31(2): 
187-199.

Mishra, A.K. and Sandretto, C.L. (2002). Stability of farm income and role of nonfarm 
income in U.S. agriculture. Review of Agricultural Economics 24(1): 208-221.

OECD (2003). Farm Household Income. Issues and Policy Responses. OECD Publications 
Service, Paris (France).

OECD (2009). Managing risk in agriculture. A holistic approach. OECD publication, Paris.
Plewa, F.J. Jr., and Friedlob, G.T. (1995). Understanding cash flow. John Wiley & Sons, 

New York (USA).
Popp, M., Rudstrom, M. and Manning, P. (2005). Spatial yield risk across region, crop and 

aggregation method. Canadian Journal of Agricultural Economics 53(1): 103-115.
Severini, S. and Tantari, A. (2013). The effect of the EU farm payments policy and its 

recent reform on farm income inequality. Journal of Policy Modeling 35(2): 212-227.
Shapiro, S.S. and Wilk, M.B. (1965). An analysis of variance test for normality (complete 

samples). Biometrika 52(3-4): 591-611.
Tangermann, S. (2011). Risk Management in Agriculture and the Future of the EU’s Com-

mon Agricultural Policy. ICTSD Paper, 34, Geneva, June.
Vrolijk, H.C.J., Poppe K.J. (2008). Income variability and income crisis in the European 

Union. In: Meuwissem, M.P.M., van Asseldonk, M.A.P.M. and Huirne, R.B.M. (eds), 
Income stabilisation in European agriculture. Wageningen Academic Publishers, 
Wageningen, The Netherlands, pp. 33-54.



79The instability of farm income

Vrolijk, H.C.J., de Bont, C.J.A.M., van der Veen, H.B. and Poppe, K.J. (2009). Variability 
of farm incomes, prices and yields in the European Union. Report 2009-005. LEI 
Wageningen UR, The Hague. 

Appendix

Table A1. Comparison of the distributions of the farms in the considered sub-sample and in the whole 
FADN sample among geographical areas and altimetry zones.

Similari
ty^

N. of 
farms %

N. of 
farms %

Geographical areas: 90%
Center 343 14% 2098 19%
Islands 171 7% 1172 10%
South 626 26% 2903 26%
Northwest 765 32% 2423 22%
Northest 499 21% 2593 23%

Whole country 2404 100% 11189 100%
Altimetry zones: 98%

Hilly 1131 47% 5072 45%
Mountain 478 20% 2326 21%
Plan 795 33% 3791 34%

Whole country 2404 100% 11189 100%

Sub-sample Whole sample 
(2012)

^ Finger and Kreinin (1979) similarity index. 100% complete similarity.
Source: own elaboration on Italian FADN data.



80 S. Severini, A. Tantari, G. Di Tommaso

Table A2. Comparison of the distributions of the farms in the sub-sample and in the whole FADN sam-
ple among types of farming and economic size.

N. of 
farms %

N. of 
farms %

Types of Farming (TF) Code 93%
Specialist field crops 1 572 24% 3007 27%
Specialist horticulture 2 276 11% 824 7%
Specialist permanent crops 3 715 30% 3073 27%
Specialist grazing livestock 4 493 21% 2504 22%
Specialist granivore 5 84 3% 524 5%
Mixed cropping 6 161 7% 691 6%
Mixed livestock and Mixed 
crops-livestock

7 103 4% 566 5%

Economic size (ESU classes) 98%
Small (Classes 1, 2 and 3) 699 29% 3100 28%
Medium (Classes 4, 5, 6) 1595 66% 7311 65%
Large (Classes 7 and 8) 110 5% 778 7%

Whole farms 2404 100% 11189 100%

Sub-sample Whole sample 
(2012) Similari

ty^

^ Finger and Kreinin  (1979) similarity index. 100% complete similarity.
Source: own elaboration on Italian FADN data.

Table A3. Evolution of farm income level by types of farming in the analysed period (2003=100).

2003 2004 2005 2006 2007 2008 2009 2010 2011 2012
Types of Farming (TF): TF

Specialist field crops 1 100 101 102 99 131 138 124 115 115 113
Specialist horticulture 2 100 97 96 99 100 108 103 101 95 73
Specialist permanent Crops 3 100 90 94 92 99 123 99 133 113 113
Specialist grazing livestock 4 100 87 90 109 107 114 104 109 112 105
Specialist granivore 5 100 137 104 79 74 66 88 76 56 74
Mixed cropping 6 100 96 91 98 140 114 123 140 145 128
Mixed livestock and Mixed 
crops-livestock 7 100 93 115 112 154 127 131 113 99 117

Total sample 100 99 97 98 106 112 105 110 103 101

Farm Income level (2003=100)

Source: own elaboration on Italian FADN data.



81The instability of farm income

Figure A1. Evolution of agricultural product deflated price indexes in the years 2003 - 2012 
(2005=100).

100

105

110

115

120

125

130

2003 2004 2005 2006 2007 2008 2009 2010 2011 2012

All	 crops

Crop	prod.s	but	vegetables

Livestock	and	 livestock	prod.s

All	agric.	products

Source: Italian National Institute of Statistics (ISTAT); www.istat.it. 

Figure A2. Evolution of agricultural input deflated price indexes in the period 2003 - 2012 (2005=100).

95

100

105

110

115

120

125

130

135

2003 2004 2005 2006 2007 2008 2009 2010 2011 2012

Intermediate	 Consumptions

Feed

Investments

General	 index

Source: Italian National Institute of Statistics (ISTAT); www.istat.it.