Unit VII J

WINTER 2025 Real estate Finance 105

Editor-in-Chief

Development Editor

Michael J. Seiler

Jayne K. leaSe

VOLUME 41, NUMBER 3 WINTER 2025

The Performance of Electronic Property Auctions Following the Greek Sovereign Debt Crisis: Factors Influencing the Price and Probability of Sales By Kornilios Vezyroglou and Fotios Siokis

This study investigates the factors that influence the likelihood of sale or the sale price of properties at real estate auctions. Our research is based on a unique cross-sectional dataset of

auctioned properties in Greece during 2018-2019, and a series of novel explanatory variables. This period followed a severe sovereign debt crisis that lasted for eight years and led to a significant increase in non-per- forming loans (NPLs) and property foreclosures. Our findings suggest that certain types of properties, such as tourist accommodations, ware- houses, and offices, have a higher probability of sale than other types of real estate. Also, residential properties have higher selling chances than commercial ones. Newly-built properties and assets located in the two major metropolitan areas tend to be in much greater demand. However, repeated auction attempts for a property appear to discourage bidders. Finally, banks seem to deploy different bidding strategies.

INTRODUCTION Property auctions have become an integral part of modern economies,

serving as a platform for enforcement proceedings or as a meeting place for willing buyers and sellers. With the growing popularity of auctions, academic research has analyzed various aspects of their mechanisms. For instance, studies have compared the efficiency of traditional sales and

Kornilios Vezyroglou, CFA, is with the Department of Balkan, Slavic & Oriental Studies at the University of Macedonia in Thessaloniki, Greece. He may be contacted at [email protected].

Fotios Siokis, PhD, is with the Department of Balkan, Slavic & Oriental Studies at the University of Macedonia in Thessaloniki, Greece. He may be contacted at fsiokis@ uom.edu.gr.

The authors declare there is no conflict of interests in this research.

The Performance of Electronic Property Auctions

106 Real estate Finance WINTER 2025

auctions (Mayer, 1995; Dotzour, Moorhead, and Winkler, 1998), as well as different auction formats (Hansen, 1985; Miller, 2014). Other research has explored the role of real estate agents in the expansion of auctions (Maher, 1989), applied behavioral finance theory to auctions (Grum and Grum, 2015), and investigated legal concerns over auctions (Good and Hammond, 2006). Despite the importance of empirical evidence in real

estate research, most authors limit themselves to developing a theoretical framework to support their arguments. This can be attributed to the scarcity of available and reliable data, which is further compounded by the unique char- acteristics of the real estate market (Belke and Keil, 2018). For instance, Hu et al. (2022) focus on the implications of property heterogeneity on coefficient estimates in statisti- cal models. Real estate also is considered an illiquid asset class due to its immobility, which makes it difficult to buy and sell quickly and easily. Illiquidity can cause problems in property valuation (Zheng and Hui, 2016), with serious implications on auction performance. Another challenging feature of real estate is the lack of transparency, meaning that the seller might hide information regarding the property’s status, leading to relatively inefficient markets (Levitt and Syverson, 2008). This study aims to shed light on the unexplored Greek

real estate auction market and examine the factors that affect the probability of a successful auction outcome, as well as the determinants of the final sale price. The study utilizes a novel and rich dataset, compiled for the first time from the top four systemic banks, and introduces a broader range of different types of real estate compared to previous work. It also employs a series of new explanatory variables, such as construction year and judicial adjustment of opening bid. Moreover, the sample includes properties from around the country and not just a specified region.1 The most influential articles that have inspired our research are those authored by DeBoer, Conrad, and McNamara (1992), Ong, Lusht, and Mak (2005), Wong et al. (2017), and Stevenson and Young (2015). To the best of our knowledge, there is no bibliography on

the auction process in Greece after the two legal amend- ments (Laws 4335/2015 and 4512/2018) that shaped its current form, making this the first attempt to provide empirical evidence. The academic community and mar- ket practitioners have always shared a common interest in understanding the factors that affect the probability of sale or sale price in real estate auctions. An insight into their

mechanism could facilitate a deeper understanding of the complex Greek case by detecting and correcting possible flaws or by distinguishing between the most and the least attractive property types, with positive influence on market effectiveness. Property auctions in Greece are a special issue due to the prolonged sovereign debt crisis and the imple- mentation of corrective measures of harsh austerity, that triggered an unparalleled spike in non-performing loans (NPLs) and consequently a substantial increase in real estate auctions. Furthermore, each auction type exhibits its own idiosyncratic features, making it essential to recognize that drawing conclusions from other auction formats may yield inaccurate messages. In 2017 NPLs peaked, accounting for 45.7 percent of the total domestic gross loans.

AUCTION SYSTEM IN GREECE FOR FORECLOSED REAL ESTATE The decade-long crisis in the Greek economy that fol-

lowed the 2007 peak led to an unprecedented decline in the domestic real estate market, with housing prices dropping by almost 42 percent until 2018. Since then, the property market in Greece has been recovering. The “Golden Visa” program has been one of the main drivers. In the first half of 2023, foreign citizens invested over €1 billion in property purchases to obtain “Golden Visa” status, nearly three times higher than the corresponding period of the previous year (€361 million). The property sector also has been positively impacted by tourism, as hotel chains in Greece doubled from 2018 to 2023. In 2022, investments in Greek commercial real estate surged by 40 percent to reach a record high of €1.65 billion. Meanwhile, apartment prices rose at an average annual rate of 11.1 percent in 2022, compared to an average increase of 7.6 percent in 2021. Foreign Direct Investments in Greek real estate also saw a significant increase of 68 percent, reaching €1.97 billion in 2022. These figures indicate a strong growth trend in the Greek real estate market, which is expected to continue in the coming years.2

Auctions in Greece have never been a mainstream channel for real estate transactions of non-distressed properties. They are used almost exclusively for selling foreclosed real estate. Until 2018, the compulsory auction process suffered from deficiencies that discouraged most potential bidders from participating. However, the Greek real estate auction market underwent two crucial developments in recent years. First, Law 4335/2015, enacted on January 1, 2016, drastically

The Performance of Electronic Property Auctions

WINTER 2025 Real estate Finance 107

limited the auction proceedings for creditors with general privileges (Tax Authorities, Social Security Funds, etc.), who until then were fully satisfied to the detriment of all other claims. Now, banks holding encumbered assets are entitled to a minimum of 65 percent of the sales proceed- ings (see Exhibit 1). Second, Law 4512/2018, enacted on February 21, 2018, mandated that all compulsory auctions could only be realized electronically thereafter. The main reason for switching to online auctions is that until 2017, auctions on foreclosed real estate were conducted physically, at magistrates’ courts in a sealed-bid formula. This process led to a tumultuous period of suspensions prompted by social groups protesting outside the courtrooms. Against this backdrop, digitalization offered easy access to every interested bidder. The two legal amendments described above increased

bidding activity, but the results were not as expected. During 2018, almost four out of five successfully auctioned properties were bought by the banks, without any bids placed from a third party. Additionally, nearly 30% of all auctions planned for 2018 were suspended or cancelled. Similar statistics apply for 2019. From 2020 until October 2023, 148,000 real estate auctions have been conducted in Greece, with a total value of 18.3 billion euros and an aver- age property value of 128,000 euros. However, only 19,200 auctions have been successful, which corresponds to 12.9 percent of the total. This period includes the years of the pandemic, when auctions were limited. Auctions under Law 4512/2018 follow the English type,3

through an open outcry ascending dynamic process. After registering on the platform, candidate bidders are required to submit an application for the auction they are interested in by physical mail, along with a receipt of the participa- tion guarantee, equal to 30 percent of the opening bid. No other participation fees, preparation costs, or security

deposits are needed. During the auction process, each can- didate can place successive bids with the opening bid as the starting point. The opening bid equals the property’s appraised commercial value, calculated by an expert of the bailiff ’s choice out of a list of certified appraisers held by the Ministry of Finance. Every new bid must exceed the existing one by at least one euro. The latest bid entered is directly visible to all participants, so that everyone knows their current personal ranking. The names of the other participants are not revealed, though. The highest final bid corresponds to the winning bidder. After full payment of the bid and the legalization of any planning violations, the successful tenderer becomes the new owner of the prop- erty, free of any pre-notations, mortgages, and attachments. However, any third party is allowed to file a replevin action within five years after the asset is transcribed in the relevant public registry (ar. 1020). The first step in the enforcement of monetary claims

entails the composition of the “writ of attachment” by a bailiff, which is delivered to the debtor and then tran- scribed in the corresponding land registry or cadaster. After the attachment, any kind of disposal of the prop- erty by the debtor is considered invalid. Within 45 days from the attachment, the borrower/current owner can file an opposition. The hearing takes place within 60 days from the request, and the court decision should be issued within 60 days. In parallel, the borrower also can request the suspension of execution until the final verdict on the opposition. Moreover, the auctioneer or the borrower can file for the upward correction of the opening price if their own assessed value differs significantly. Third parties also are entitled to contest the enforcement proceeding during all its phases, as long as their legal rights are infringed. The auction date is set by law between seven and eight

months from the publication date to provide enough time

Exhibit 1—Distribution of ProcEEDings

Types of auction claims Types of claims announced at an auction

A & B & C A & B B & C A & C

A. General Privileges [ranked by order of seniority] Claims for wages – legal fees – claims of State for Value Added Tax – claims of Social Security Funds – claims of State and other public entities for any other reason

25% 1/3 (not announced) 70%

B. Special Privileges [ranked by order of priority] Maintenance expenses – mortgage-backed claims

65% 2/3 90% (not announced)

C. Claims without collateral 10% (not announced) 10% 30%

The Performance of Electronic Property Auctions

108 Real estate Finance WINTER 2025

for the aforementioned legal procedures. If the auction is declared fruitless, the creditor who initiated the auction is asked if they want to acquire the property. Otherwise, a new auction is scheduled within forty days. According to the legislation in force during the period examined, if there was no bidding again, any creditor could file for a reduc- tion of the opening bid by providing the court with a new appraisal completed after the second fruitless auction. The court decides whether or not to lower the opening bid or even order the divestment of the asset to a creditor or a third party at a certain price (a rare case).

DAY AFTER BANK REPOSSESSION Real estate is not considered a core business banking

activity, although banks are de facto major real estate players. Banking institutions occupy properties for operational rea- sons, repossess failed foreclosures (REOs), finance directly (loans to real estate developers, hotels, etc.), or hold them as collateral for the provision of loans of any type. Regarding REOs, several attempts have been made to depict the pos- sible benefits from active management (Lewis and Webb, 2007; Scardovi, 2016). Still, as a rule, banks try to minimize their REO portfolio on both business and legal grounds. A prolonged holding period of REOs jeopardizes banks’

financial health in various ways. The first obvious short- coming is price risk, since REOs tend to sell at a signifi- cantly lower price than the gross loan value at foreclosure. Liquidity risk is also a factor in managing inventories of buildings, particularly in times of sluggish real estate mar- kets. Banks are also faced with operational risk, in terms of compliance with a complex and ever-changing regulatory framework, which falls outside their area of expertise. Last but not least, the repossession of properties involves reputa- tion risk, particularly when the weakest social classes are hit. Even if a bank could see some potential advantages in

keeping REOs for a longer period, current legislation in both Europe and the United States offers no room for discussion. In Europe, the European Banking Authority (EBA) issued its latest guidelines on the management of non-performing and forborne exposures in October 2018. According to Article 229, credit institutions should classify repossessed assets as held for immediate sale in their present condition (IFRS 5.7). The proposed timeframe for sale is set at one year, during which active marketing policies must be pursued (IFRS 5.8). Although EBA guidelines are not legally binding, members are expected to comply. As for the

United States, pursuant to the Code of Federal Regulations Title 12 § 34.82, national banks are obliged to dispose of REOs at the earliest time that prudent judgment dictates but not later than the end of the holding period (or an extension thereof) permitted by law. The permitted holding period is five years. The socioeconomic drawbacks of prolonged REO keep-

ing are not to be ignored, as evidenced by the Great Recession. The foreclosing crisis following the 2007- 2009 financial crash peaked in 2010, to a point that total mortgage debt surpassed US national GDP. The inevitable large-scale accumulation of REOs from banks brought widespread repercussions, including value reduction of nearby homes and a rise in criminality. Sumell (2009) and Ihlanfeldt and Mayock (2014) provide novel evidence about the persistence of negative neighborhood externalities, even after REOs are sold and reenter the market. Gould, Madar, and Weselcouch (2014) stress the need for more drastic policy intervention to address the high REO concentra- tion in certain regions, in the logic of the Neighborhood Stabilization Program.4

Unsurprisingly, banks actively pursue the divestiture of their REO properties. To facilitate this process, they have established online platforms to showcase these properties.5 The options presented to investors range from direct sales to participation in English-type online auctions. Because this auctioning procedure is not mandated by law, REO owners are not bound by the legislation outlined in the previous section. Nevertheless, it’s noteworthy that these voluntary auctions closely resemble their compulsory counterparts.6 Only a small portion of total REOs is available for rent. Additionally, a series of properties is accessible through the servicers that have acquired distressed loans.7

LITERATURE REVIEW A considerable body of literature exists on different aspects

of the real estate auction process, although most studies are based on theoretical models. Few studies have managed to process real data from past auctions to analyze the factors that influence the auction sale probability and/or the final sale price. DeBoer, Conrad, and McNamara (1992) set the ground-

work by examining the expected determinants of the probability of sale and selling price in English ascending bid auctions of tax delinquent properties. They found that residential properties (apartments and single-family homes) have better selling chances than vacant land, residential or

The Performance of Electronic Property Auctions

WINTER 2025 Real estate Finance 109

commercial. Residential properties tend to sell at higher prices, too, although no statistical significance was detected. Properties located downtown had fewer selling chances and at lower prices than others. Finally, the assessed market value proved a significantly positive determinant of both the probability of sale and the sale price. It should be noted, though, that the authors assumed that auction participants assess market value in the same way, without sufficient jus- tification. Moreover, the pool of determinants tested was limited. Ong, Lusht, and Mak (2005) examined the factors that

affect auction sale probability in English ascending bid auc- tions and added new variables, such as bidder turnout and the auctioning house. Both factors proved positive and sta- tistically significant. Stevenson and Young (2015) also pro- vided evidence that in English-type real estate auctions, the number of bidders/bids is generally significant not only in the probability of sale at auction but also with respect to the sale price achieved. Besides, Ong, Lusht, and Mak (2005) were the first to work on a mixed sample of distressed and non-distressed properties. The former were shown to sell more easily, in alignment with the popular belief that their sellers are more anxious to recoup part of their investment. Additionally, the state of the market proved statistically sig- nificant, supporting the expectation that depressed market conditions can lead to lower demand. Wong et al. (2017) expanded the empirical research on

the forces that drive sale probability and sale price in the real estate auction market. By including several new fac- tors, such as seller ethnicity, number of online viewers, and previous auction attempts, they found that a high-income ethnicity group is more likely to acquire a property and at a higher price. Superior performance might also be indica- tive of an ethnic group’s desire to reinforce its presence in the area. These observations are truly interesting in light of the ever-growing multicultural character of modern metro- politan cities. Finally, past unsuccessful attempts increase the chance of sale. Indeed, in most auction systems, real estate assets tend to be offered at a lower price after a series of failed attempts. Empirical studies collectively share some similar findings,

including the critical role of location factors, as auction theory suggested (Asabere and Huffman, 1992). They also confirm the theory that a higher reserve price8 normally leads to fewer selling chances but to a higher final sale price.9 However, with the limited number of empirical studies, drawing generalized conclusions carries a high

degree of risk. The diversity of auction systems adds to the problem. For example, the auction procedure in Sweden, empirically tested by Hungria-Gunnelin (2013), is a ver- sion of the original English type with some unique features, which hinder a meaningful comparison to other formats. Moreover, a closer look at the literature reveals several other problematic areas. For instance, the methodology followed suggests a difficulty in quantifying psychological factors in decision making. However, these factors can radically alter bidders’ a priori perception of the asset value, especially when it comes to online auctions, as Ariely and Simonson (2003) note. The state of maintenance of auctioned proper- ties is also overlooked, despite its obvious importance.

DATA AND DESIGN Our cross-sectional sample consists of 4,168 data points,

representing auctions of properties of various types for the period 2018-2019, covering all of Greece. Inflation has not been taken into account, as it stayed at extremely low levels during these two years (0.63 percent and 0.25 percent respectively). The data were provided by the Retail Collections Unit of a major systemic Greek bank. The ini- tial dataset included 11,149 transactions, but entries with missing data were excluded. This data can be considered “missing completely at random” (MCAR). In this case, it is generally considered safe to exclude the data from the analysis. Besides, filling missing data with mean or imputed values can lead to biased results, since it is usually based on fragile assumptions (Rubin, 1996; Schafer, 1997; Little, 1988). Additionally, we have excluded unique real estate types that appear only a few times from our dataset, lead- ing to a less heterogeneous sample. Including them in the analysis could have a disproportionate impact on the results, making it difficult to detect a true cause-effect relationship. Finally, repeated failed auction attempts on the same asset were excluded, so each auction corresponds to a distinct property, and only the latest attempt is considered. Given that our sample consists of thousands of assets, with some of them auctioned up to seven times, keeping discrete val- ues for the same property would probably yield misleading results. Despite the significant reductions, the final sample still includes a substantial amount of data for statistical analysis. In fact, most studies in the field, presented in the literature review above, were conducted with much smaller samples. Our empirical analysis is carried out on two axes. First, we

explore the factors that contribute to a successful auction

The Performance of Electronic Property Auctions

110 Real estate Finance WINTER 2025

by utilizing a probit model. Later, we trace the determinants of the final sale price by employing an OLS regression. The following equation describes the probit model, where the dependent is a dummy variable (OUT) that takes a value of 1 in case of a successful auction outcome and 0 otherwise. Approximately 61 percent of all sampled properties were bid on and sold.10

OUT = f (LN_BID, AUC_BANK, MAIN_M, LAND_M, RES, RES_W, PARK, COM, STORE, TOUR, OFF, OTHER_RE, YEAR, CITY, ATT, V_CH) (1)

The equation (2) represents the multiple linear regres- sion model, where the natural logarithm of the sale price (LN_SALE) is the dependent variable. Assets that were not sold are assigned a value of zero.

LN_SALE = f (AUC_BANK, A, B, C, D, MAIN_M, LAND_M, RES, RES_W, PARK, COM, STORE, TOUR, OFF, OTHER_RE, YEAR, CITY, ATT, V_CH) (2)

The input variables of the models differ in two aspects. Firstly, the natural logarithm of the opening bid (LN_BID) appears only in equation (1), as it is highly correlated with the sale price (LN_SALE) (0.998), given that usually there was only one active bidder. Secondly, equation (2) is enriched with four dummy variables, corresponding to the four systemic banks (A, B, C, D), that collectively control more than 95 percent of the domestic market.11 Each bank placing the highest bid is assigned a value of 1. The omitted fifth variable WINNER_OTH equals 1 if the winner is an individual or a firm other than a bank.12 This dummy vari- able is redundant because it can be inferred from the other four variables A, B, C, and D. Before deploying our analysis, we should mention that

previous research in similar cases has accounted for the presence of sample selection bias by applying a Heckman two-stage process (DeBoer, Conrad, and McNamara, 1992; Wong et al., 2017). This method has gained considerable popularity over time in various fields, albeit often with- out sufficient justification. In our case, the exceptionally large sample imposes the use of a relatively low level of significance, under the assumption of indifference between Type I and Type II errors (Kim, 2015). Indeed, the inverse Mills ratio (λ), which is considered the main indicator of

sample selection bias, was found statistically significant at a 0.01 level of significance. However, we rejected its use after taking into account two criteria proposed by Certo et al. (2016): (1) we found that LN_BID, the so-called exclu- sion restriction to omit in the second phase of the process, was not significant in the first stage (the probit model) at any level of significance, even 0.10, and (2) the correlation between LN_BID and λ was trivial. Exhibit 2 presents descriptive statistics of the sample.

Similar to the sale price (LN_SALE), the opening bid is expressed as a natural logarithm (LN_BID) to mitigate extreme skewness. As long as the opening bid is lower than an investor’s reserve price, it can be viewed as a buying opportunity. The variable AUC_BANK is a dummy vari- able and equals 1 when the auctioneer is a banking institu- tion. As stated before, in Greece, auctions are inextricably linked to distressed sales that are mainly held by banks. The variables MAIN_M and LAND_M stand for main space and land size, respectively (both in square meters). The vari- ables RES, RES_W, PARK, COM, STORE, TOUR, OFF, LAND are also dummy variables defining the type of prop- erty. They denote residential properties of all sorts, residential storage space, parking space, commercial property, storage, tourist accommodations, office space, and land, respectively. Land (LAND) and residential properties (RES) sum up to around 60 percent of all assets sold. This is reasonable, as these kinds of real estate are the ones mainly encumbered in retail banking. Like WINNER_OTH before, LAND is the omitted dummy variable regarding property type, as including all dummy variables related to property type in the model would result in multicollinearity. The variable OTHER_RE consists of buildings of mixed

use or property rights. The variable YEAR is a time dummy variable assigned the value of 1 if the property was con- structed after January 1, 2001. At that time, the Antiseismic Regulation was enforced since Greece is one of the most earthquake-prone countries in the world. The building’s age, as a loose approximation of quality, is expected to be a prominent selection criterion among bidders. Variable CITY is assigned the value of 1 if assets come from the metropolitan areas of the Prefectures of Attica (where the capital Athens is situated) and Thessaloniki (the second largest city). The geographical dispersion of the auctioned properties in our sample is almost equally split between these two areas and the rest of Greece. As explained above, we chose to match each value with a

unique asset. Dummy variables V_CH and ATT reflect all

The Performance of Electronic Property Auctions

WINTER 2025 Real estate Finance 111

Exhibit 2—DEscriPtivE statistics

Variable Representation Min Mean Max Std. Dev.a Description Auction outcome OUT 0 0.610 1 (0.488) Dummy variable = 1 if property was

sold at auction Natural logarithm of sale price

LN_SALE 6.120 10.607 15.010 (1.486) Continuous variable

Natural logarithm of opening bid

LN_BID 6.120 10.723 15.425 (1.422) Continuous variable

Property auctioned by bank

AUC_BANK 0 0.951 1 (0.217) Dummy variable = 1 if property was set by a bank

Main space occu- pied (sq. meters)

MAIN_M 0 114.613 5,144.500 (246.771) Continuous variable

Plot area occupied (sq. meters)

LAND_M 26,740 1,730.430 67,500.000 (3,725.090) Continuous variable

Residential property

RES 0 0.465 1 (0.499) Dummy variable = 1 for detached houses or apartments

Land LAND 0 0.135 1 (0.342) Dummy variable = 1 for land Office OFF 0 0.037 1 (0.189) Dummy variable = 1 for offices/office

buildings Residential warehouse

RES_W 0 0.088 1 (0.284) Dummy variable = 1 for residential warehouses

Store STORE 0 0.129 1 (0.335) Dummy variable = 1 for stores/shop- ping centers

Commercial building

COM 0 0.050 1 (0.218) Dummy variable = 1 for commercial buildings

Tourist facility TOUR 0 0.006 1 (0.074) Dummy variable = 1 for hotels or other tourist establishments

Parking space/ station

PARK 0 0.070 1 (0.256) Dummy variable = 1 for spaces/build- ings related to parking

Other types of real estate

OTHER_RE 0 0.020 1 (0.141) Dummy variable = 1 for all remaining types of real estate

Year of construction

YEAR 0 0.408 1 (0.492) Dummy variable = 1 if property was built in 2001 [land gets a value of 0]

City CITY 0 0.441 1 (0.497) Dummy variable = 1 for properties located in the Prefectures of Athens or Thessaloniki

Change in value V_CH 0 0.054 1 (0.225) Dummy variable = 1 if the initially set opening bid was changed by court decision

Auctioning attempts

ATT 0 0.963 1 (0.189) Dummy variable = 1 if the property has been auctioned once or twice

Bank A A 0 0.044 1 (0.206) Dummy variable = 1 if the winning bidder is Bank A

Bank B B 0 0.383 1 (0.486) Dummy variable = 1 if the winning bidder is Bank B

Bank C C 0 0.037 1 (0.188) Dummy variable = 1 if the winning bidder is Bank C

Bank D D 0 0.060 1 (0.238) Dummy variable = 1 if the winning bidder is Bank D

Winner other than Banks A-B-C-D

WINNER_OTH 0 0.476 1 (0.499) Dummy variable = 1 if the winning bidder is a natural person or a firm other than bank

a Standard deviations in parentheses.

The Performance of Electronic Property Auctions

112 Real estate Finance WINTER 2025

the missing information resulting from the exclusion. The variable V_CH indicates whether the initial opening bid has been modified due to a petition for upward price correc- tion, before the auction takes place. The variable ATT splits auctioning attempts into two categories. Value 1 means that the property has been auctioned once or twice.13 It would be useful to calculate how consecutive auction attempts affect the final possibility of a successful outcome. This information could be valuable to auctioneers, so that they revisit their strategy after the first failed round. More specifically, 91 percent of all successfully auctioned proper- ties were sold on the first attempt, suggesting that further efforts bring questionable results. Finally, concerning A, B, C, and D, we plan to check if all banks play by similar rules or each one follows a distinct bidding policy. At this point, it should be stressed that in our case banks don’t act as agents for other sellers but as sellers of assets collateral- ized by themselves. Thus, any comparisons with previous research on the influence of auctioning houses should be carried out prudently. The literature reviewed above loosely suggests several

hypotheses of statistically significant relationships between the independent and dependent variables of the probit model. Specifically, we could hypothesize that:

A. An increase in the opening bid, as expressed by vari- able V_CH, is related to lower selling chances.

B. The type of auctioneer (AUC_BANK) affects the probability of sale.

C. Properties auctioned more than twice (ATT) sell easier.

Hypothesis A is derived from the findings of DeBoer, Conrad, and McNamara (1992), who concluded on a negative relationship between opening bid and selling probability. Thus, we should expect a negative coefficient for V_CH. Hypothesis B is based on the work of Ong, Lusht, and Mak (2005), who spotted different bidding behavior among auctioning houses. Finally, hypothesis C comes from Wong et al. (2017), who found that the number of previous auction attempts is positively related to sale probability. It should be mentioned that features such as bidder turn-

out and reserve price, which have been examined in the past, are absent in our dataset. At the same time, our database includes various property features that are not included in previous empirical studies. As a result, we were unable to

derive hypotheses on the OLS regression based on existing literature.

EMPIRICAL RESULTS The empirical results are presented in Exhibit 3. For

the probit model, the area under the Receiver Operating Characteristic (ROC) curve, which plots sensitivity against 1-specificity, is 0.7275,14 a score that signals fairly good diagnostic accuracy. Given that probit coefficients do not have an intuitive meaning per se, we present the marginal effect of each variable as well. With regard to the OLS model diagnostics, no collinearity is detected, but there is evidence of heteroscedasticity, which is remedied by re- running the regression with robust (Huber-White) standard errors. The final OLS model explains about 71.76 percent of the variance of the natural log of the observed final prices, as expressed by adjusted R-squared. The coefficient of the variable V_CH is significantly

negative, verifying hypothesis A: a rise in the opening bid lowers bidding interest. Concerning hypothesis B, a bank’s decision to issue an auction program (AUC_BANK) is statistically significant for both the outcome and the price achieved. Real estate assets have additional 10.79 pecent selling chances when auctioned by financial institutions, an expected outcome, as Greek banks are the principal actors in the domestic auction market and have an obvi- ous motive to discard the accumulated NPLs. The posi- tive coefficient in the probit model reflects their frequent willingness to bid on their own auctioned property.15 Bank B seems to show a preference for cheaper proper- ties compared to A and D. The high p-value of bank D leaves no opportunity for safe conclusions about its bid- ding policy. Auctioned items with smaller main space (MAIN_M)

tend to sell more easily, since they cost less than larger ones. Residential properties sell almost twice as easily as com- mercial ones, when comparing the two coefficients. The coefficient of land plot size (LAND_M) approaches zero in both equations, and has a high p-value in the OLS regres- sion. In fact, the few experimental approaches up to now are controversial: Ooi, Sirmans, and Turnbull (2006) report land size as a significant positive factor for land price, while Amidu and Agboola (2009) outline that lot size has a statis- tically insignificant impact on the price premium paid for auctioned residential properties. All real estate categories exhibit statistically significant

coefficients in both equations, except for OTHER_RE

The Performance of Electronic Property Auctions

WINTER 2025 Real estate Finance 113

in the probit equation. Tourist facilities (TOUR) have the highest marginal effect on the probability of sale, highlight- ing the dominant role of tourism in the Greek economy. Residential properties (RES, RES_W) generally sell easier than commercial ones (COM), a sign of an economy rav- aged by prolonged austerity measures during the previous decade. Residential warehouses (RES_W) and parking spaces (PARK) often are auctioned together with the apart- ments that they are contractually attached to, so their almost identical coefficients in the probit regression are expected.16 Offices (OFF) are bidders’ second most common choice. The years 2018-2019 were characterized by domestic growth, and the correlation between GDP and office occu- pancy cycles already has been noticed (for example, Mueller

and Peiser, 2015). Commercial buildings (COM) and stores (STORE) are the least appealing acquisitions. Our probit model shows that the coefficient LN_BID is

positive, but not statistically significant, in contrast to earlier findings about the negative effect of the opening bid on the probability of sale. Sales of newer buildings exhibit higher demand than older ones (YEAR), but the link between age and selling price remains obscure. Bidders show a prefer- ence for properties in Athens and Thessaloniki (CITY), and better selling chances lead to higher prices. A striking result is that repeated offers of the same asset (ATT) may reduce the probability of sale17 (hypothesis C rejected). A possible explanation is that consecutive failed attempts are viewed as a sign of low-quality or even a dysfunctional property

Exhibit 3—Probit anD oLs rEgrEssion rEsuLts

Explanatory Variable

Probit Model OLS Model

Estimates of probit Marginal effect analysis

Coefficient St. error Coefficient St. error Coefficient St. error

LN_BID 0.0245 (0.0233) 0.0094 (0.008) -

AUC_BANK 0.2755*** (0.0934) 0.1079*** 0.3710 0.5707*** (0.1001)

MAIN_M -0.0009*** (0.0002) -0.0003*** 0.0001 0.0040*** 0.0004

LAND_M 0.0000*** (0.0000) 0.0000*** 0.0000 0.0000 0.0000

RES 0.5841*** (0.0719) 0.2185*** 0.0262 -0.3336*** 0.0915

RES_W 0.8331*** (0.1203) 0.2657*** 0.0293 -3,2756.0000*** 0.0996

PARK 0.8362*** (0.1212) 0.2640*** 0.0287 -2,5687.0000*** 0.0889

COM 0.2901** (0.1290) 0.1052** 0.0440 -1,7448.0000*** 0.1636

STORE 0.2517*** (0.0838) 0.0927*** 0.0297 -0.5017*** 0.1003

TOUR 1.1662*** (0.2949) 0.3123*** 0.0414 -1,1550.0000** 0.5379

OFF 0.9089*** (0.1351) 0.2750*** 0.0282 -1,3015.0000*** 0.1144

OTHER_RE 0.2122 (0.1559) 0.0779 0.0548 -1,3539.0000*** 0.3130

YEAR 0.2278*** (0.0466) 0.0861*** 0.1740 0.0370 0.0332

CITY 0.4176*** (0.0443) 0.1570*** 0.0163 0.1713*** 0.0340

ATT 0.9249*** (0.1183) 0.3536*** 0.0398 0.1817 0.1517

V_CH -0.1746* (0.0931) -0.0679* 0.0367 0.1975** 0.0770

BANK A - - 0.4429*** 0.0788

BANK B - - 0.2708*** 0.0568

BANK C - - 0.4185*** 0.0912

BANK D - - 0.0041 0.0855

CONS -1.7282*** (0.2865) - 10,1532.0000*** 0.1882

Standard errors in parentheses.

*** Significant at the 1% level, ** 5% and * 10% respectively. ‘CONS’ stands for the constant term.

The Performance of Electronic Property Auctions

114 Real estate Finance WINTER 2025

because bidders have no access to technical or legal assess- ments nor the opportunity to inspect the auctioned prop- erty as the ownership still belonged to the delinquent borrower. Thus, investors’ interest gradually fades. In the OLS equation, the ATT coefficient is positive but statis- tically insignificant. This finding concurs favorably with those of Wong et al. (2017), who found that the statistical significance of auctioning attempts was eliminated once they restated the initial dependent variable (the natural log of sales price) as a ratio per square feet. To challenge our results, we modified Equation (2) in

two ways. First, we focused only on the metropolitan areas of Athens and Thessaloniki, whose increased signifi- cance to investors already has been identified in the initial

regressions. By placing variable CITY as the dependent dummy variable, we decreased our initial sample to 1,328 observations. Secondly, we examined only assets which were built after 2001, where the dummy variable YEAR is assigned the value of 1 and the sample consists of 1,185 observations. We ran a Breusch-Pagan / Cook-Weisberg test, which revealed the presence of heteroscedasticity. The problem was tackled with the use of robust standard errors. The results of the experimental setup are presented in Exhibit 4. Regarding the first robustness check, all coefficients exhibit

the same sign and status of statistical significance as in the initial setting, thus corroborating our prior findings. It is worth noting that now tourist facilities (TOUR) have a

Exhibit 4—robustnEss chEcks

Explanatory Variable OLS Regression Model Results for Athens and Thessaloniki (CITY=1)

OLS Regression Model Results for buildings after 2001 (YEAR=1)

Coefficient St. error Coefficient St. error

AUC_BANK 0.5563*** 0.1309 0.5319*** 0.1542

MAIN_M 0.0050*** 0.0010 0.0039*** 0.0005

LAND_M 0.0000 0.0000 0.0000 0.0000

RES -0.8370*** 0.1690 -0.2368 0.1616

RES_W -3.8337*** 0.1449 -3.3794*** 0.1982

PARK -2.9499*** 0.1359 -2.4477*** 0.1899

COM -2.5159*** 0.2289 -1.6609*** 0.1997

STORE -1.1161*** 0.1687 -0.4065 0.1726

TOUR -9.0899*** 2.0545 -0.3211 0.8355

OFF -1.9429*** 0.1777 -0.9697*** 0.1838

OTHER_RE -1.9659*** 0.5461 - -

YEAR 0.0537 0.0439 - -

CITY - - 0.2108*** 0.0429

ATT 0.0117 0.1532 0.2221 0.2404

V_CH 0.3277*** 0.0887 0.2822** 0.1139

BANK A 0.2658** 0.1132 0.3784*** 0.1091

BANK B 0.2215*** 0.0645 0.1909** 0.0812

BANK C 0.3617*** 0.1340 0.2935 0.1841

BANK D -0.0290 0.1136 -0.0033 0.1136

CONS 10.9587*** 0.2239 10.1238*** 0.3650

Standard errors in parentheses.

*** Significant at the 1% level, ** 5% and * 10% respectively.

‘CONS’ stands for the constant term.

Variable ‘OTHER_RE’ excluded because of multicollinearity.

The Performance of Electronic Property Auctions

WINTER 2025 Real estate Finance 115

significantly lower coefficient value, indicating that most of the successfully auctioned luxurious resorts were located in the rest of Greece. In addition, increases in opening bids by court decision (V_CH) are more intense, given that proper- ties in the two largest Greek cities are much more expensive. As for the second OLS version with properties built after

2001, a main differentiation from the original sample is that three variables become statistically insignificant: residential properties (RES), tourist establishments (TOUR), and win- ning bids from Bank C ©. The superior value of auctioned properties in the two largest cities (CITY) is once again validated. The V_CH variable can be interpreted as in the first check. Lastly, we can confirm our observation that Bank A is willing to spend more money to acquire a property compared to B, presumably because of a larger stockpile of NPLs to eliminate.

DISCUSSION—CONCLUSIONS The aim of this study was to provide an innovative pre-

sentation of the Greek real estate auction market. The rigor- ous statistical analysis adds to the limited relevant academic research. At the same time, it offers useful insights to all participants (prospective bidders, auctioneers, regulatory authorities, etc.) about the dynamics that shape the Greek real estate auction market. The considerable sample size and time span, combined with the diversity of the property types, allows us to generalize our conclusions to the whole Greek real estate auction market. Our findings become even more interesting in light of the fact that the NPL ratio is still substantially above the EU average, inflow of new NPLs remains contained but risks persist,18 and activity in the auc- tion market of distressed properties is expected to persist in the following years. Consequently, we believe that the aim of the study has been achieved. Hypothesis A about the negative relationship between

opening bid and sale probability is supported by the data, as evidenced by the statistically significant negative coefficient of V_CH: whenever the opening bid was adjusted upwards, the selling chances were fewer. Our findings are also con- sistent with hypothesis B, which posited that the type of auctioneer affects the probability of sale; items auctioned by banks proved to sell easier and at higher prices com- pared to others. Finally, the results of this study did not pro- vide support for hypothesis C: Contrary to previous papers, we found that if an asset is auctioned again and again, its probability of sale fades, probably because repeated failures carry a negative connotation about the property’s status.

Moving beyond the hypotheses, our analysis reveals that not all banks follow the same bidding philosophy, as some show a tendency towards more expensive acquisitions. Since tourism represents the country’s comparative advan- tage, tourist facilities attract the majority of the winning bids. As expected, real estate in Athens and Thessaloniki is more sought-after and thus more expensive. The same applies to properties built after 2001. Addressing Greek banks’ asset quality has been repeat-

edly stated as a top priority for the ECB. One of the incentives introduced to offer assistance in dealing with the NPL backlog was the mandatory electronic holding of auctions, since February 2018. Limited investing interest from third parties forced banks to incur the cost of their non-performing exposures. The main problems related to this inefficient procedure were: (1) the significantly high opening bid, based on the initial valuation or even an upward reassessment requested in court by the borrower; (2) the slow judicial process to apply for a reduced open- ing bid after two consecutive fruitless attempts, combined with frequent rejections of those requests; (3) the limited information offered on the auctioned properties; and (4) the abuse of certain legal actions by strategic defaulters, especially the application to court just before the auction in house insolvency proceedings under the 3860/2010 (Katseli) Law on indebted households, leading to a sus- pension of enforcement measures for a period usually exceeding two years.19 Any corrective measure to combat these deficiencies would normally help to intensify bid- ding activity. Further investigations could extend the conclusions drawn

here, as the Greek auction market remains terra incognita with the exception of our article. Research could focus on the two largest metropolitan areas or on a specific type of real estate. Interesting questions might also derive from a new database including information that was not available in the present study, such as the property’s state of mainte- nance, or the reserve price set by the auctioneer.

REFERENCES Amidu, A. R., & AGBOOLA, A. O. (2009) Empirical evi-

dence of the influences on first-price bid auction premi- ums, International Real Estate Review, 12(2), pp. 157-170.

Ariely, D. and Simonson, I. (2003) Buying, Bidding, Playing, or Competing? Value Assessment and Decision Dynamics in Online Auctions, Journal of Consumer Psychology, 13(1-2), pp. 113–123.

The Performance of Electronic Property Auctions

116 Real estate Finance WINTER 2025

Asabere, P. K., and Huffman, F. E. (1992) Price determi- nants of foreclosed urban land, Urban Studies, 29(5), pp. 701-707.

Belke, A. and Keil, J. (2018), Fundamental Determinants of Real Estate Prices: A Panel Study of German Regions, International Advances in Economic Research, 24(1), pp. 25-45.

Certo, S.T., Busenbark, J.R., Woo, H. and Semadeni, M. (2016) Sample selection bias and Heckman models in strategic management research, Strategic Management Journal, 37(13), pp. 2639–2657.

Dovles, A. (2019) Enforcement of Domestic Judgments in Civil and Commercial Matters, Greek Law Digest vol. 2.

DeBoer, L., Conrad, J. and McNamara, K.T. (1992) Property Tax Auction Sales, Land Economics, 68(1), p. 72.

Dotzour, M., Moorhead, E. and Winkler, D. (1998) The Impact of Auctions on Residential Sales Prices in New Zealand, Journal of Real Estate Research, 16(1), pp. 57–72.

Eauction.Gr (2023), Candidate Bidder Instructions Manual. Retrieved from https://www.eauction.gr/Manuals/ IL.SY.PLEIS%20-%20Candidate%20Bidder%20 Instructions%20Manual%20-%20eauction.gr.pdf.

Enterprise Greece (2023) Real Estate 2023. Retrieved from https://www.enterprisegreece.gov.gr/assets/content/files/c27/ a8492/f230/RealEstate2023.pdf.

European Banking Authority (2018) Guidelines on man- agement of non-performing and forborne exposures, European Banking Authority, 31 Oct. 2018, https://www. eba.europa.eu/regulation-and-policy/credit-risk/guidelines-on- management-of-non-performing-and-forborne-exposures.

European Commission (2019) Enhanced Surveillance Report, February 2019. Retrieved from https://economy- finance.ec.europa.eu/system/files/2019-02/ip019_en.pdf.

European Commission (2023) Post-Programme Surveillance Report on Greece, Spring 2023. Retrieved from https://economy-finance.ec.europa.eu/system/files/2023-05/ ip203_en.pdf.

Grum, B. and Grum, D.K. (2015) A model of real estate and psychological factors in decision-making to buy real estate, Urbani izziv, 26(1), pp. 82–91.

Gould, E.I., Madar, J. and Weselcouch, M. (2014) The Foreclosure Crisis and Community Development: Exploring REO Dynamics in Hard-Hit Neighborhoods, Housing Policy Debate, 24(1), pp. 1-28.

Hammond, C.M. and Good, S.L. (2006) Real Estate Auctions: Legal Concerns for an Increasingly Preferred

Method of Selling Real Property, The Real Property, Probate and Trust Journal, Vol. 40 No. 4, 2005-2006, pp. 58-83.

Hansen, R. (1985) Empirical Testing of Auction Theory, The American Economic Review, 75(2), 156-159.

Hu, Y., Lu, B., Ge, Y. and Dong, G. (2022) Uncovering spatial heterogeneity in real estate prices via combined hierarchical linear model and geographically weighted regression, Environment and Planning B, vol. 49(6), pp. 1715-1740.

Hungria-Gunnelin, R. (2013) Impact of number of bidders on sale price of auctioned condominium apartments in Stockholm, International Real Estate Review, 16(3), pp. 274-295.

Ihlanfeldt, K. R. and Mayock, T. (2014) The Persistence of Segregation in the Metropolis: New Findings from the 2010 Census, Journal of Urban Economics, 84, pp. 34-45.

Kim, J.H. (2015) How to Choose the Level of Significance: A Pedagogical Note, SSRN Electronic Journal.

Levitt, S.D. and Syverson, C. (2008) Market Distortions When Agents Are Better Informed: The Value of Information in Real Estate Transactions, Review of Economics and Statistics, 90(4), pp. 599–611.

Lewis, D. and Webb, J.R. (2007) Potential cost synergies from banks acquiring real estate brokerage services, Journal of Banking & Finance, 31(8), pp. 2347-2363.

Little, R.J. (1988). A test of missing completely at random for multivariate data with missing values, Journal of the American Statistical Association, 83 (404), pp. 1198-1202.

Maher, C. (1989) Information, Intermediaries and Sales Strategy in an Urban Housing Market: The Implications of Real Estate Auctions in Melbourne, Urban Studies, 26(5), pp. 495–509.

Mayer, C.J. (1995) A Model of Negotiated Sales Applied to Real Estate Auctions, Journal of Urban Economics, 38(1), pp. 1–22.

Miller, J.J. (2014) From English to First-Price Sealed Bid: An Empirical Assessment of the Change in Auction Type on Experienced Bidders, Review of Economic Perspectives, 14(2), pp. 105–127.

Mueller, G., & Peiser, R. (2015) Real estate cycles and their strategic implications for investors and portfolio man- agers, Journal of Real Estate Portfolio Management, 21(1), pp. 1-13.

Ong, S.E., Lusht, K., Mak, C.Y. (2005) Factors influencing auction outcomes: Bidder turnout, auction houses and

The Performance of Electronic Property Auctions

WINTER 2025 Real estate Finance 117

market conditions, Journal of Real Estate Research, 27(2), pp. 177–192.

Ooi, J. T. L., Sirmans, C. F., & Turnbull, G. K. (2006) The option value of vacant land, SSRN Electronic Journal.

Rubin, D. B. (1996) Multiple imputation after 18+ years, Journal of the American Statistical Association, 91 (434), pp. 473-489.

Scardovi, C. (2016) Holistic Active Management of Non-Performing Loans, SpringerLink, doi: 10.1007/978-3-319-25363-3.

Schafer, J. L. (1997) Analysis of Incomplete Multivariate Data, CRC Press, doi: 10.1201/9780367803025.

Stevenson, S. and Young, J. (2015) The Role of Undisclosed Reserves in English Open Outcry Auctions, Real Estate Economics, 43(2), pp. 375–402.

Sumell, A. J. (2009), The Persistence of Negative Neighborhood Effects: Evidence from the Gautreaux Program, Journal of Human Resources, 44(2), pp. 303-332.

Tzortzi, E. (2023) The unlikely estate agents. Kathimerini. Retrieved from https://www.ekathimerini.com/ economy/1211150/the-unlikely-estate-agents/.

U.S. Department of Housing and Urban Development (2023) Neighborhood Stabilization Program. Retrieved from https://www.hud.gov/program_offices/comm_planning/nsp.

Wong, W.-C., Lee, J.Y.M., Daud, M.N. and Ng, P.L. (2017) Seller ethnicity and property characteristics: foreclosure sales in Malaysia, International Journal of Housing Markets and Analysis, 10(4), pp. 539–551.

Zheng, X. and Hui, E.C.M. (2016) Does liquidity affect housing market performance? An empirical study with spatial panel approach, Land Use Policy, 56, pp. 189-196.

NOTES 1. Singapore has been examined as a whole in previous papers; however, it is virtually a city

state.

2. Data retrieved from the 2023 real estate factsheet issued by Enterprise Greece, the official investment and trade promotion agency of the Greek State. Available at https://www. enterprisegreece.gov.gr/assets/content/files/c27/a8492/f230/RealEstate2023.pdf.

3. For a comprehensive presentation of the Greek auction system, the reader is referred to Dovles (2019) and to the official candidate bidder instructions manual, available at www. eauction.gr.

4. The Neighborhood Stabilization Program (NSP) was activated three times with a view to offering emergency assistance to communities with high rates of abandoned or foreclosed residences, and especially to households with annual incomes up to 120 percent of the area median income (AMI). Source: https://www.hud.gov/program_offices/ comm_planning/nsp.

5. The websites are: realestateonline.gr (National Bank); findyourproperty.gr (Eurobank); properties4sale.gr (Piraeus); and astikaakinita.gr & propertynow.gr (Alpha).

6. Further analysis of the auction processes followed by the banks on their platforms exceeds the scope of the present paper.

7. Servicers, such as Intrum and DoValue, are anticipated to offer approximately 9,600 properties within 2024. See https://www.ekathimerini.com/economy/1211150/ the-unlikely-estate-agents/.

8. Reserve price can be defined as the minimum price that a seller would be willing to accept.

9. This observation is rejected only by Ong, Lusht and Mak (2005) but probably due to the proxy used to gauge reserve price.

10. Given that banks almost always reveal their auction performance as a total, and not per Collection Unit, the verification of our findings is practically impossible. The problem is intensified by the reporting inconsistencies among banks, for example, some of them report number of auctions and others report number of auctioned items. Only one of them offers detailed information per Division in its 2018 report. Their official announcement regarding the recovery rate of the Retail Unit (about 60 percent) fits almost perfectly with our results (about 61 percent).

11. The banks are Eurobank, Piraeus Bank, National Bank of Greece and Alpha Bank (in random order). The mean value of Bank B (Exhibit 2) is high because B is the supplier of the data.

12. It is worth mentioning that in the initially received dataset of 11,149 observations, the Tax Authorities appeared only 43 times as issuers of an auction program, and never got to acquire an asset. Also, the winning bidder was a smaller, non-systemic bank in only four cases of our sample. These data were eliminated for computational reasons.

13. It is clarified that there is no overlap between variables V_CH and ATT—their correla- tion is only 0.11.

14. Sensitivity: the ratio of correctly predicted real estate assets that sold at tax auction, that is, OUT=1. Specificity: the ratio of correctly predicted assets that didn’t sell, that is, OUT=0. In our model, sensitivity is 58.58 percent and specificity equals 75.09 percent. The cutoff point was calculated at 0.67.

15. According to their own reports, banks acquired approximately 80 percent of the assets that they auctioned.

16. It is noted that variable PARK contains only a few parking station enterprises. It is mainly about parking spaces, which are in legal terms distinct properties but are practi- cally linked to certain housing properties within the same plot.

17. As previously stated, after the second fruitless auction, any involved creditor can petition the court for a reduced opening bid.

18. Source: European Commission Post-Programme Surveillance Report on Greece, Spring 2023. Available at https://economy-finance.ec.europa.eu/system/files/2023-05/ip203_ en.pdf.

19. See European Commission, Enhanced Surveillance Report, February 2019.

Copyright of Real Estate Finance (Aspen Publishers Inc.) is the property of Wolters Kluwer Legal & Regulatory and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.