搜档网
当前位置:搜档网 › Information asymmetry and self-selection bias in bank loan announcement studies

Information asymmetry and self-selection bias in bank loan announcement studies

Information asymmetry and self-selection bias in bank loan announcement studies
Information asymmetry and self-selection bias in bank loan announcement studies

Information asymmetry and self-selection bias in bank loan announcement studies$

Pankaj K.Maskara a,Donald J.Mullineaux b,n

a Eastern Kentucky University,United States

b University of Kentucky,United States

a r t i c l e i n f o

Article history:

Received4February2010

Received in revised form

9August2010

Accepted2September2010

Available online30March2011

JEL classi?cation:

G14

G21

Keywords:

Loans

Announcements

Information asymmetry

Event studies

Selection bias

a b s t r a c t

Event-study driven research has produced a consensus that loans are unique relative to

other?nancial contracts.But these studies assume that small samples of loan

announcements adequately represent the loan population.We?nd that loan announce-

ments are rare and driven by factors such as information asymmetry and perceived

materiality.We show that the sample used by Billett,Flannery,and Gar?nkel(1995)

fails to represent the loan universe and that signi?cant abnormal announcement

returns are con?ned to their smallest?rms.Our sample,which better represents the

loan population,produces an abnormal return insigni?cantly different from zero.The

?ndings suggest that self-selection bias affects extant loan announcement research and

do not support the views that loans are a special form of?nance or that private and

public debt differ in signi?cant ways.Were all loans to be announced,the average

abnormal return would likely be insigni?cant.

&2011Elsevier B.V.All rights reserved.

1.Introduction

There is consensus in the literature that bank loans

differ signi?cantly from other forms of?nance.That

conclusion follows from a large body of research showing

that announcements of bank loan agreements result in

positive abnormal equity returns for borrowers,on average.

Because announcements of bond?nancings generate no

signi?cant abnormal returns and stock issues yield nega-

tive abnormal returns,researchers infer that loans are

unique or special.Although banks originate a large number

of commercial loans every year,loan announcement

studies typically use relatively small samples.1Nonethe-

less,researchers commonly draw inferences from the

results concerning the population of loans,because they

implicitly assume that all loans are equally likely to be

announced.

We focus on the prospect that loan announcements

may be selective and investigate the factors that affect the

decision to announce a loan.We also examine how

sample selection problems can bias the outcomes of loan

announcement studies.We?nd that only about one-

fourth of bank loan acquisitions are announced and that

?rms with announced loans differ systematically from

those with unannounced loans.The presence of

Contents lists available at ScienceDirect

journal homepage:https://www.sodocs.net/doc/6e880725.html,/locate/jfec

Journal of Financial Economics

0304-405X/$-see front matter&2011Elsevier B.V.All rights reserved.

doi:10.1016/j.j?neco.2011.03.019

$The authors thank Paul Childs,Zekeriya Eser,John Garen,Yuri

Khoroshilov,Brad Jordan,Susan Jordan,Debarshi Nandy,Joe Peek,

Robert Schweitzer,and Hassan Tehranian,with a special thanks to an

anonymous referee,for very helpful comments and suggestions.We also

thank Rajesh Maskara for technical support,Jackie Thompson for

editorial suggestions,and Anita Maskara for excellent research

assistance.

n Corresponding author.

E-mail address:mullinea@https://www.sodocs.net/doc/6e880725.html,(D.J.Mullineaux).

1The average sample size is446,with a range from117to728,in

the studies by James(1987),Lummer and McConnell(1989),Slovin,

Johnson,and Glascock(1992),Best and Zhang(1993),Preece and

Mullineaux(1994),and Billett,Flannery,and Gar?nkel(1995).

Journal of Financial Economics101(2011)684–694

information asymmetries strongly affects the prospect that a borrower will announce a loan.Material loans, meaning those that are large relative to the borrowing ?rm’s asset base,or loans made to?rms facing actual or prospective cash?ow problems,are also more likely to be announced.Loans that remain unannounced go to large, less information-problematic?rms where investors are less likely to view a new loan as material information.We conjecture that loans to these?rms would not elicit a signi?cant market response,even if made public.Among announced loans,we?nd that only those to the smallest decile of borrowers elicit signi?cant reaction.Conse-quently,we conclude that only a small subset of loans is special,at best.

The paper proceeds as follows.Section2offers hypoth-eses about the nature of bank loan announcements, addressing issues the literature has ignored to date,such as what parties will announce loans and under what circumstances.Section3offers several tests of the hypotheses,and provides estimates of logistic models for loan announcements.Section4provides an analysis of the results of a well-known study of loan announce-ments motivated by our?ndings on announcement decisions.Section5concludes.

2.Why do borrowers–or others–announce loans?

The conclusion that bank loans are a special type of ?nancial contract and/or that banks have unique capabil-ities relative to other?nancial?rms follows from research based on event studies.Many researchers report that positive and signi?cant abnormal returns attend to announcements that?rms have signed a bank loan agreement.James(1987)is a widely cited example of such research and he?nds a sizeable average excess return of193basis points.He views the result as con-sistent with Fama’s(1985)conjecture that banks are ‘‘unique’’institutions because they gain insider-like infor-mation through lending and deposit relationships.Many studies replicate the results of James’s paper,although often with quali?cations.2For instance,Lummer and McConnell(1989),Preece and Mullineaux(1994),and Billett,Flannery,and Gar?nkel(1995)report signi?cant excess returns of0.61%,0.79%,and0.68%,respectively. Generally,researchers seem to agree that loans differ somehow from other debt contracts and that institutions offering loans are likewise unique,in some sense.

The relevant population for addressing whether loans differ from other?nancings is the totality of loans made to business borrowers.However,unlike other forms of ?nance raised by public companies,the Securities and Exchange Commission(SEC)does not generally require the public announcement of loan?nancings.3The excep-tion to this rule would be a loan‘‘that arises other than in the ordinary course of doing business’’(SEC,Form8-K, General Instructions).Nor does the SEC consider a bank loan a‘‘security’’potentially subject to registration requirements.Consequently,a?rm’s decision to reveal a bank loan acquisition is in most cases discretionary.The extant literature therefore necessarily relies on a sample of loans that get announced,with the implicit assumption that announced and unannounced loans are identical in all important respects.We examine this assumption’s validity by considering the factors that would motivate a borrower to reveal the successful conclusion of a loan agreement.Corporate?nance decisions are seldom ran-dom and managers typically self select their choices in a wide variety of situations(Chaney,Jeter,and Shivakumar, 2004;Li and Prabhala,2007;McNichols and O’Brien, 1997).Firms may have preferences not just across differ-ent types of funding,but also about whether to publicize their?nancing decisions.

Announcing a loan can lower information asymmetry between a borrowing?rm and its investors.Diamond (1985)demonstrates that disclosure can be an optimal policy both because it generates cost savings for investors who would otherwise attempt to acquire costly informa-tion and because it can improve risk sharing by making investor expectations more homogeneous and reducing the speculative positions of informed traders.Diamond and Verrecchia(1991)show that increased disclosure can decrease a?rm’s cost of capital by reducing information asymmetries.But the sizeable disclosure literature reveals a large variation in predicted outcomes.For example, Verrecchia(2001)notes that,in some models,increased disclosure results in more information asymmetry and that the empirical evidence is sparse on the relationship between disclosure and information asymmetries.Our initial hypothesis directly links loan disclosures to infor-mation asymmetries.

Hypothesis1.Firms that present higher levels of infor-mation asymmetry to investors will be more likely to announce their loan?nancings.

Although SEC rules do not mandate the reporting of bank loans,public?rms are required to disclose all ‘‘material’’events within four business days of their occurrence on Form8-K(SEC Staff Accounting Bulletin No.99—Materiality).Accounting rules state an event is material if it potentially impacts the?nancial position of the?rm or the value of its shares.We hypothesize that borrowers will view bank loans as material in certain circumstances and consequently will be more inclined to reveal such?nancings.For example,if a bank loan is large relative to the borrower’s existing asset base,the new

2For example,loans generate positive abnormal returns and con-sequently are special when they are made(1)to small?rms(Slovin, Johnson,and Glascock,1992),(2)to?rms facing earnings uncertainty (Best and Zhang,1993),(3)by nonbanks(Preece and Mullineaux,1994), (4)by reputable lenders(Billett,Flannery,and Gar?nkel,1995),(5)by syndicates with few lenders(Preece and Mullineaux,1996),(6)in amounts of$10billion or more(Mosebach,1999),(7)by lenders not using loan sales or securitization(Marsh,2006),or(8)with larger portions retained by arrangers(Focarelli,Pozzolo,and Casolaro,2008).

Billett,Flannery,and Gar?nkel(2006)?nd that bank loans are not special,however,when abnormal returns are estimated over a longer period,such as three years.

3Public issues of debt or equity must be disclosed via the registra-tion process.Private issues of equity,debt,or capital and operating leases must be disclosed in an8-K?ling.

P.K.Maskara,D.J.Mullineaux/Journal of Financial Economics101(2011)684–694685

debt is more likely to affect the?rm’s?nancial position and thus qualify as material.And?rms with cash?ow problems may view a new bank loan as relevant to its ?nancial position or equity value.To assess these claims, we test a second hypothesis.

Hypothesis 2.A borrower will be more likely to announce a bank loan when the?nancing is a large component of the capital structure or when the?rm faces higher prospects of cash?ow problems.

Although borrowing?rms have an option to announce the successful completion of a loan agreement,they are not the sole potential source of such information.Lenders and reporters with?nancial newspapers or information services(such as Bloomberg or Reuters)also could reveal such?nancings.But we argue that these alternative sources are less likely to be motivated by efforts to reduce information asymmetries at the borrowing company.This permits the testing of a third hypothesis that allows us to assess the relevance of information asymmetries from an alternative perspective.

Hypothesis3.Loan announcements generated by parties other than the borrower will be more weakly related to ?rm information asymmetries than will borrower-gener-ated announcements.

3.Testing the hypotheses

3.1.The sample

To examine our hypotheses empirically,we need information on unannounced as well as announced loans. Loan announcement studies typically generate a database of bank loan announcements over some chosen sample period and then estimate the equity market’s response to the announcements.In particular,researchers perform keyword searches in news databases such as the Wall Street Journal and Dow Jones Newswire,looking for key-words such as credit,loan,commitment,or line of credit. By design,this methodology captures only loans that are announced in the media during the study period.We employ a different methodology.Rather than begin with loan announcements,we start with a sample of randomly selected loans from the population and then search for announcements.We consider a loan to be announced if any article in the Factiva database mentions the loan.Our population is the set of loans to U.S.borrowers that appears in the Reuters Loan Pricing Corporation’s(LPC) DealScan database from1987to2004.The data in Deal-Scan primarily come from SEC?lings,large loan syndica-tors,and a staff of reporters.The database is widely used in studies of various aspects of the loan market.4We delete borrowers that are government entities,?nancial companies,or utilities and observations for?rms lacking a ticker symbol,leaving20,140loans.We then randomly pick200loans without replacement from this population. We analyze the characteristics of this sample relative to those of the population and?nd it representative.The characteristics include loan amount and maturity,lender identity,the number of facilities in the loan,borrower rating,and the year the loan is originated.We next search for announcements of these loans in the Factiva database, which collects its information from10,000authoritative sources,including the Wall Street Journal,Financial Times, and the news services of Dow Jones,Reuters,and the Associated Press.Rather than use a computer program to search for announcements,we read the text of news stories and search for each announcement manually to minimize measurement error.Our search window is six months prior to and two months after the loan closing date.

We?nd some mention in the media for57of the200 loans in our initial sample.Of these instances,?ve media reports show a borrower is either seeking the loan or expecting to receive the loan or that a lead bank is inviting syndicate members to participate in a loan.The remaining52announcements con?rm that the loan is made.Of these,the company generates37,and the lending bank only one.We treat a loan as company-announced when the media states‘‘in a press release the company said,’’or‘‘the company announced today.’’We also treat an announcement as company-generated when top management contributes a quote to the news story. Either reporters or SEC?lings are the source of informa-tion in ten announcements.We cannot identify the information source in four announcements.Of the52 closed deals,one announcement takes place17days before closing.The others are within15days of closing, with seven on the day the loan closed and12on the next trading day.Eleven loan announcements are made prior to the loan closing date.We treat loans announced after 4:00PM EST as announced on the next trading day.

With one exception,the announcements in our?rst sample take place within15days of the loan closing date, so we next randomly draw600additional loans from our population and search for announcements in a narrower window of715days from the loan closing date.Again, we must con?rm that our sample of loans is representa-tive of the population.In Table1,we show the average loan and borrower characteristics of our sample loans are not statistically or economically different from those of the loans in the DealScan population.

Once more,a little more than a quarter of the600 loans are announced.Most loans continue to go unan-nounced.Fig.1shows the histogram for the168of the 232announcements that are generated by the borrowing ?rm.These announcements typically occur within one to seven days of the loan closing date.The most popular

4As one example,Gande and Saunders(2006)use the LPC database in their analysis of whether banks remain special following the advent of the secondary loan market.The authors do not employ the usual method of searching for loan announcements.Instead,they note the database

contains several dates for most loans,including the signing date,active date,and closing date.They take the earliest of these dates as the likely announcement date.In effect,they assume that all loans in the database (footnote continued)

are announced and predict the likely announcement date.As our results will show,this assumption appears to be untenable.

P.K.Maskara,D.J.Mullineaux/Journal of Financial Economics101(2011)684–694 686

announcement date is the day after the loan closes,but18 of these loans are announced prior to closing.The histo-gram for all232announcements(not shown)is similar to that of Fig.1.One difference is that borrowers are less likely than others to announce loans before the closing date.

3.2.The relevance of information asymmetry

We now analyze the likelihood of loan announcement, with a view toward testing our?rst hypothesis—that information asymmetries will be a systematic driver of borrowers’loan announcement decisions.We get data on borrowing?rms from the Center for Research in Security Prices(CRSP),Compustat,Institutional Brokers’Estimate System(I/B/E/S),and the SEC’s Ownership Reporting Sys-tem(ORS)data sets.The literature employs several mea-sures of information asymmetry,including market microstructure-based variables(bid-ask spread,turnover), analysts’earnings per share(EPS)forecasts(analyst fol-lowing,forecast error,dispersion of analyst opinion),the volatility of stock returns(volatility of abnormal returns around earnings announcements,residual volatility),and the intensity of insider trading.For example,Gomes and Phillips(2010)use a variety of measures to study how information asymmetry affects a?rm’s decision to issue private versus public securities,including analysts’forecast errors,the dispersion of analyst opinion,the volatility of abnormal returns around earnings announcements,insider trading,and a composite index based on stock returns and turnover.Bharath,Pasquariello,and Wu(2009)analyze the impact of information asymmetry on capital structure decisions,employing a composite measure based on mar-ket microstructure variables.Leary and Roberts(2010) likewise study the information asymmetry-capital struc-ture relationship,using analyst dispersion,analyst follow-ing,?rm size,and?rm age as proxies for information asymmetry.

The literature lacks a?rm consensus on the optimal measure(s)of information asymmetry,so we create a composite index based on six common information asym-metry benchmarks—analyst forecast errors,dispersion of analyst opinions,volatility of residual returns,volatility of abnormal returns around earnings announcements,?rm age,and bid-ask spreads.5Forecast error is the absolute difference between analysts’predicted earnings and actual earnings per share in the month prior to the annual earnings announcement.Dispersion of analyst opinions is the standard deviation of analysts’forecasts of annual EPS in the last month prior to the earnings announcement. Both the forecast error and volatility variables are stan-dardized by share price.The volatility of abnormal returns around earnings announcements is the standard deviation of three-day abnormal returns around earnings reports in the?ve-year period preceding loan announcements.We use the CRSP value-weighted index to compute market-adjusted abnormal returns and require at least ten obser-vations in the?ve-year period.Residual volatility in daily stock returns is the standard deviation of market-adjusted daily stock returns in the year of the loan announcement. Following Chung and Zhang(2009),we measure the bid-ask spread variable as the average ratio of the difference between the daily bid and ask closing prices to the midpoint of the bid and ask closing prices.6We require at least100observations per year to calculate spread and residual volatility.Firm age is measured as the number of years since the?rst?rm observation in Compustat and serves as an inverse proxy for information asymmetry. We create the information asymmetry index by calculat-ing the six measures of information asymmetry for our sample?rms for each year in our sample period.To avoid

Table1

Descriptive statistics for the loan population in the Dealscan database

and for our randomly selected sample of loans.

Panel A shows DealScan database descriptive statistics for all loan

deals between1987and2004involving U.S.borrowers that are not

government entities,?nancial companies,or utilities.Debt rating is the

Standard and Poor’s(S&P)credit rating for the borrower’s senior debt at

the time of the loan acquisition.It takes a value of one for C-rated

borrowers,two for those rated CC,three for CCC-rated and so on.

Number of lenders represents the number of lenders participating in

the loan deal.Tenor measures loan maturity in months.Deal amount is

the loan amount in millions of dollars.Interest rate is the credit spread

on the loan(measured as the loan rate minus the London Interbank

Offered Rate(LIBOR)in basis points.Syndicated is a dummy variable

that takes a unit value for loan deals that had two or more lenders

participating and zero otherwise.Panel B shows similar descriptive

statistics for the randomly selected800loans in our sample.

Variable N Mean Std.

dev.

Minimum Maximum

Panel A:Population

Debt rating6,949 5.51 1.319

Number of

lenders

20,127 5.467.71110

Tenor18,01144.78341366

Deal amount20,1402737080.0525,000

Interest rate16,234199.68137à141,490

Syndicated20,1270.590.4901

Panel B:Sample

Debt rating342 5.69 1.1839

Number of

lenders

800 6.158.831108

Tenor71541.4632.720361

Deal amount800309.6719.850.212,000

Interest rate676182129.38 6.32980

Syndicated8000.630.4801

5We de?ne and measure our information asymmetry variables

following Dierkens(1991),Krishnaswami,Spindt,and Subramaniam

(1999),Krishnaswami and Subramaniam(1999),Bharath,Pasquariello,

and Wu(2009),Gomes and Phillips(2010),and Leary and Roberts

(2010).

6We also estimate the bid-ask spread using daily high and low

prices and the relatively complex methodology of Corwin and Schultz

(2010)and?nd our results to be marginally stronger(results unre-

ported).We also use a Trades and Quotes(TAQ)database measure and

the private information(PIN)measure of Easley,Hvidkjaer,and O’Hara

(2010)and again?nd our results to be marginally stronger.This

measure is available at https://www.sodocs.net/doc/6e880725.html,/site/hvidkjaer/data for

all NYSE/Amex common stocks from1983to2001.PIN,the perceived

probability of the arrival of an informed trade,is one of the components

of the information asymmetry index created by Bharath,Pasquariello,

and Wu(2009).They report that PIN has the highest correlation of all

the components with their information asymmetry index.We also?nd

signi?cant positive correlation between PIN and our information asym-

metry index measure(Table2).

P.K.Maskara,D.J.Mullineaux/Journal of Financial Economics101(2011)684–694687

the effect of secular trends in the measures over time,we group ?rms into quintiles based on each measure for all the ?rms in the year a loan is announced.7As Gomes and Phillips (2010)do,we compute the information asymmetry index as the average of the quintile ranking of a ?rm based on the six information asymmetry

measures.

102030405060<-1

-1

1

2

3

4

5

6

7

>7

A n n o u n c e m e n t s

Business days after loan start date

Fig.1.The timing of loan announcements by borrowing ?rms.A histogram of the number of loans announced in the media by the borrowing ?rm on the loan closing date (Day 0)and certain days before and after.Our randomly selected sample of 800loans from the DealScan database yielded 232announced loans.We consider a loan to be announced if any news story in the Factiva database mentioned it explicitly.Of the 232announcements,the borrowing ?rm made 168.We treat the borrower as the source if the news story stated that the company announced the ?nancing or if a company representative contributed a quote to the story.The sample period is 1987–2004.

Table 2

The Pearson correlation coef?cients across different measures of information asymmetry for our sample ?rms.

Firm size is the log of the market capitalization of the borrowing ?rm.Bid-ask spread is the average ratio of the difference between daily closing bid and ask prices to the midpoint of closing bid and ask prices.Forecast error is the ratio of the average absolute difference between analysts’EPS forecasts and actual EPS to the share price of the ?rms.Forecast dispersion is the standard deviation of the analyst annual EPS forecasts divided by ?rm’s share price.Analysts’forecasts are observed at the last month prior to annual earnings announcements by ?rms and share price as of the beginning of the month prior to earnings announcements.Volatility around earnings announcements is the standard deviation of three-day market-adjusted abnormal returns for the last ?ve-year period.Residual volatility is the standard deviation of the daily market-adjusted abnormal return of the ?rm’s stock.Insider trading intensity is the ratio of total yearly purchases and sales of a ?rm’s stock by of?cers,management,and directors to annual trading https://www.sodocs.net/doc/6e880725.html,rmed trade is the perceived probability of arrival of informed trades (PIN)as calculated by Easley,Hvidkjaer,and O’Hara (2010).Firm age is the number of years since the ?rst observation for the ?rm in Compustat.Number of analysts is measured as number of analyst EPS estimates in the last month prior to earnings reports.Turnover is the average ratio of the number of shares traded daily divided by the number of shares outstanding.The information asymmetry index is the average quintile ranking of a sample ?rm in the year of the loan based on six measures of information asymmetry —forecast error,forecast dispersion,residual volatility,volatility around earnings announcements,?rm age,and the bid-ask spread.The numbers in bold re?ect signi?cance at the 1%level.The italicized numbers re?ect signi?cance at the 5%level.

IA

Firm size

Bid-ask

Forecast error

Forecast dispersion

Volatility around

earnings

Residual volatility

Insider trading

Informed trades

Firm age

Number of analysts

Firm size (0.69)

Bid-ask

0.30(0.33)Forecast error 0.26(0.16)0.18Forecast dispersion

0.19(0.12)0.220.92Volatility around earnings 0.66(0.44)0.200.240.10Residual volatility 0.69

(0.56)

0.51

0.280.190.63Insider trading 0.11(0.12)0.140.01(0.00)0.180.17Informed trades 0.53(0.77)0.310.130.160.330.410.17Firm age (0.54)0.51(0.10)(0.08)(0.08)(0.21)(0.27)(0.05)(0.40)Number of analysts (0.51)0.78(0.24)(0.11)(0.08)(0.28)(0.35)(0.11)(0.52)0.38Turnover

0.03

0.19

(0.12)

(0.01)

(0.02)

0.18

0.07

(0.06)

(0.21)

(0.01)

0.18

7

For example,the minimum tick size was one-eighth prior to 1997and a penny after 2001.If we were to categorize ?rms into quintiles based on bid-ask spreads for the sample ?rms in the year of the loan announcement,?rms that announced loans after 2001would tend to fall in the lowest quintile regardless of the level of information asymmetry.

(footnote continued )

To avoid this effect,we categorize ?rms into quintiles based on the appropriate measure for all sample ?rms in the same year.

P.K.Maskara,D.J.Mullineaux /Journal of Financial Economics 101(2011)684–694

688

Other candidate measures of information asymmetry exist.For example,several studies use turnover as an inverse proxy of information asymmetry.Dierkens(1991) notes that the literature fails to clarify the direction of turnover’s in?uence on information asymmetry,how-ever.8Another inverse proxy for information asymmetry is the number of analysts following a?rm,because analysts have a fundamental role in reducing information asymmetry.But Chung,McInish,Wood,and Wyhowski (1995)claim that analysts are attracted to?rms with more severe information problems because private infor-mation’s value increases with information asymmetry. If so,analyst following could be positively related to information asymmetry.Our results are robust to includ-ing analyst following as an inverse proxy of information asymmetry in our index.Some studies use measures of a ?rm’s growth opportunities,like the market-to-book and price-to-earnings ratios,as proxies for information asym-metry.Clarke and Shastri(2001)empirically examine the quality of information asymmetry measures and state that proxies based on?rm investment and growth opportunities are indirect measures of information asym-metry,at best.Bharath,Pasquariello,and Wu.(2009) construct an alternative information asymmetry measure based on the intensity of insider trading activity,but note that the data in the SEC’s ORS are available only between 1978and2000.Our results again hold if we include this measure in our index.We present the correlation matrix of the different information asymmetry measures in Table2.Our information asymmetry index is signi?cantly correlated with all the proxies employed in the literature except for turnover.We?nd a correlation coef?cient of à0.69between our index and?rm size.Although analyst following and PIN are not components of our information asymmetry index,we observe high correlations between these measures and our index.

In Panel A of Table3we analyze how announcements generated by borrowers and other sources vary as the amount of information asymmetry changes.The results in column2show that as information asymmetry increases, borrowers announce a higher percentage of loans.The proportion of loans announced for?rms with information asymmetry index44is more than twice as large as that for borrowers with an index value between2and3,for example,and more than three times as large relative to ?rms with information asymmetry index r2.9These

Table3

Distributions of sample?rms and announcements for various borrower and loan characteristics.

The sample consists of800loans randomly selected from a population of20,140loans in the DealScan database over the period1987–2004.Column1 of Panels A–D shows the total number of loans in the sample for which the borrowing?rm met the designated criteria.Column2of Panels A–D shows the number of loans announced by the borrowing?rm as a percent of total loans in each category.Column3of Panels A–D shows the number of loans announced by someone other than the borrowing?rm as a percent of all loans that met the criteria.Panel A tabulates the loans based on the information asymmetry index value calculated as the average quintile value of the borrowing?rm based on six measures—analysts’forecast error,dispersion of analysts’forecasts,residual volatility of stock returns,standard deviation of abnormal returns around earnings announcement,?rm age,and bid-ask spread.No data in Panel A indicate that insuf?cient data were available in CRSP,Compustat,and I/B/E/S to calculate any of the measures needed to compute the information asymmetry index value.Panel B tabulates the loans based on the number of analysts following the borrowing?rm.No data in Panel B indicate that no data were available for the analyst forecasts of earnings for the borrowing?rms in the I/B/E/S database.Panel C tabulates the loans based on the EBITDA-to-total assets ratio of the?rm.Panel D tabulates the loans based on loan size-to-asset size of the borrowing?rm.Missing data in Panels C and D indicate that the Compustat database had no data for the borrowing?rm.

(1)(2)(3)(1)(2)(3)

N Company(%)Others(%)N Company(%)Others(%)

Panel A:Information asymmetry index Panel C:EBITDA/total assets

r0107374

IA r21261160o x o0.101611811

2o IA r32581690.10r x o0.15186229

3o IA r42272680.15r x o0.20162197

IA44132358x Z0.20119138

No data57145No data65206

Total800218Total800218

Panel B:Analyst following Panel D:Loan-to-assets ratio

x r0.0511154

1–42263080.05o x r0.1100118

5–814327120.1o x r0.25232198

9–151331250.25o x r0.5158308

16–41134510x40.51343513

No data164234No data65206

Total800218Total800218

8Dierkens(1991)?nds that information asymmetry and turnover

are positively correlated.We?nd that the turnover measure in our study

is negatively correlated with bid-ask spread and positively correlated

with volatility measures and unrelated to analysts’forecasts measures

(Table2).Our results remain the same if we include turnover as an inverse proxy for information asymmetry in our index.Turnover is the average number of shares traded daily divided by the number of shares outstanding in the year of the loan announcement.

9Data constraints prevent us from computing all six of the informa-tion asymmetry measures for every?rm.For an average?rm in our sample,the index contains?ve of the six information asymmetry measures.Our results do not change if we exclude observations where

P.K.Maskara,D.J.Mullineaux/Journal of Financial Economics101(2011)684–694689

results are consistent with our ?rst hypothesis.The ?gures in column 3of Panel A reveal no increase in the proportion of loans announced by parties other than the borrower as information asymmetry increases,which supports our third hypothesis.

We next examine how loan announcements vary with analyst following,which we treat as an inverse informa-tion asymmetry proxy.Panel B shows that the proportion of loans announced is six times larger for ?rms with four or fewer analysts as compared with borrowers with 16or more.And the percentage of loans announced declines monotonically with the scale of analyst coverage.Bor-rowers that announce loans have 4.32analysts,on aver-age,while those who fail to announce have twice as many analysts at the mean.The results likewise support our ?rst hypothesis.No discernible pattern appears between the proportion of loans announced by others and the extent of analyst following,which is consistent with the results for the information asymmetry index and again supports our third hypothesis.3.3.The relevance of other factors

We next analyze announcement decisions with a focus on measures of the extent to which a loan acquisition might be judged material information.We use the state of the borrower’s cash ?ow position and loan size relative to the size of the ?rm to gauge materiality.We use earnings before interest,taxes,depreciation,and amortization (EBITDA)/Assets as a measure of the borrower’s cash ?ow prospects and suggest that investors are more likely to view a loan as material if the borrower has negative EBITDA.We use the loan’s size as a percentage of assets as a gauge of its signi?cance in the borrower’s capital structure and posit that a loan becomes more material to investors as this ratio increases.The results in Table 3,Panel C,show that ?rms with EBITDA r 0in the year prior to the loan announce a substantially higher percentage of their loans than ?rms with EBITDA in excess of 20%of assets.But little variation appears in the percentage of borrower-announced loans for ?rms with positive cash ?ows.Because ?rms with negative EBITDA have higher prospects of ?nancial distress,obtaining a loan in the face of such conditions is likely to be material news.The percentage of loans announced by others does not vary in any systematic way with the borrower’s cash ?ow status,however.Panel D of Table 3shows that when the size of a new loan exceeds half a borrower’s existing asset base,borrowers announce 35%of their loans,but when a new loan is less than 5%of assets,companies announce only 5%of the https://www.sodocs.net/doc/6e880725.html,rger loans appear to be more material and more likely to generate a borrower announcement.In this case,the portion of loans announced by others also increases with the signi?cance of the loan relative to the size of the ?rm.The results in

Panels C and D support our second hypothesis,that borrowers are more likely to announce loans that repre-sent potential material information to shareholders.

In Table 4,we disaggregate our sample into loans announced by the borrowing ?rms,those announced by others,and unannounced loans.We compute mean values for borrower and loan characteristics,including the infor-mation asymmetry index,market capitalization,the new-loan-to-assets ratio,a dummy equal to one for negative EBITDA,and the loan amount.We also perform t -tests to determine whether the observed differences are statisti-cally signi?cant.Firms with borrower-announced loans have higher information asymmetry index values,are smaller,and are more likely to have negative EBITDA than ?rms with unannounced loans,and each difference is signi?cant.Although borrower-announced loans are smaller,on average,than unannounced loans,they form a signi?cantly larger portion of the company’s asset base.The results again are consistent with our ?rst two hypotheses:information asymmetry is a differentiating factor for borrower-announced and unannounced loans,and loans announced by borrowers are more material –compose more of the ?rm’s capital structure or go to ?rms with higher prospects of cash ?ow problems –than loans that go unannounced.

3.4.A logistic model for loan announcements

We next analyze our data in a multivariate setting.We estimate coef?cients of a logit model to predict the probability a borrower will announce its loan in the

Table 4

Difference of means test for loan and borrower characteristics of announced and unannounced loans.

The sample consists of 800loans randomly selected from a population of 20,140loans in the DealScan database over the period 1987–2004.Columns 1,2,and 3show the mean values for loans announced by the borrowing ?rm,announced by any entity other than the borrowing ?rm,and unannounced loans,respectively.Column 4shows the t -statistics for difference of mean values for loans announced by the company and those https://www.sodocs.net/doc/6e880725.html,rmation asymmetry index is calculated as the average quintile value of the borrowing ?rm based on six measures–analysts’forecast error,dispersion of analysts’forecasts,residual vola-tility of stock returns,standard deviation of abnormal returns around earnings announcement,?rm age,and bid-ask spread.Firm size is the market capitalization of the borrowing ?rm.Firm size and loan amount are measured in $millions.Negative EBITDA is a dummy variable that takes unit value when the borrowing ?rm has zero or negative EBITDA,zero otherwise.Loan-to-assets is the ratio of the loan amount to the size of the existing asset base of the borrowing ?rm.

(1)(2)

(3)

(4)Announced by company

Announced by others

Unannounced t -Stat (1)–(3)Information asymmetry index 3.49

3.08

2.98

6.13

Firm size

718.912,112.605,606.58(6.35)Loan-to-assets 0.440.530.26 4.77Negative EBITDA 0.260.070.12 3.60Loan amount 191.04343.91340.90(3.61)

N

168

64

568

(footnote continued )

the information asymmetry index value contains only one or two components.We have made the information asymmetry index value for all CRSP ?rms from 1975to 2009available at http://www.pkmas https://www.sodocs.net/doc/6e880725.html,/iaindex.htm .

P.K.Maskara,D.J.Mullineaux /Journal of Financial Economics 101(2011)684–694

690

media.The dependent variable initially takes a value of one if the borrower announces its loan and zero other-wise.The explanatory variables in the model are the information asymmetry index,a dummy equal to one for?rms with negative EBITDA and zero otherwise,and the log(1tloan-size/borrower-assets).Our hypotheses suggest positive coef?cient signs for the three variables.

Column1of Table5presents the results of this estimation,and we?nd con?rming evidence for our hypotheses.Borrowers are more likely to announce loan agreements in the presence of more information asym-metry,higher prospects of cash?ow problems(negative EBITDA),and as the loan in question becomes a larger percentage of the borrower’s existing asset base.The probability is8%that a?rm with an information asym-metry index value of1,positive EBITDA,and a loan-to-assets ratio of0.30will announce its loan,but the probability increases more than fourfold–to34%–if the information asymmetry index value increases to5. Likewise,if the loan-to-assets ratio for this?rm increases from0.30to1.0,other things equal,the announcement probability increases from8%to20%.The announcement probability also increases from8%to20%if the?rm in question switches from positive to negative EBITDA.

In column2,we include the log of the borrower’s market capitalization as an inverse proxy for information asymmetry.10We?nd a highly signi?cant,negative coef-?cient estimate for the?rm size variable,while the sign and signi?cance of the remaining two variables are similar to those in column 1.Smaller?rms,which presumably present investors with more serious informa-tion asymmetry problems,are signi?cantly more likely to announce their loans.A borrower with$300million in market value,positive EBITDA,and an average loan-to-assets ratio is almost twice as likely to announce its loan as an otherwise similar?rm with$10billion in market value.

To test Hypothesis3,which posits different outcomes depending on the source of the announcement,we esti-mate a multinomial logit model.The dependent variable takes a value of one for loans announced by borrowers and a value of two for those announced by others,zero otherwise.The results appear in columns3and4of Table5.We posit that business reporters and lending institutions are less likely to announce loans based on perceptions of information asymmetry problems between ?rms and investors.This implies that no(or at best a weak)relationship should exist between the information asymmetry index and the probability of loan announce-ments by non-borrowers.11We likewise argue that the borrower’s cash?ow position is unlikely to drive a reporting decision by non-borrowers,suggesting another insigni?cant coef?cient estimate for the negative EBITDA dummy variable.But reporters and lending institutions do seem likely to report loans that are relatively large and consequently material.We therefore predict a positive coef?cient estimate for log(1tloan-size/borrower-assets).Columns3and4present the coef?cient estimates of our multinomial logit model for announcements by borrower and others,respectively.The sign and signi?-cance of the estimates in column3are similar to those in column1.But the coef?cient estimates for the informa-tion asymmetry index and negative EBITDA dummy in column4are insigni?cantly different from zero,support-ing our prediction that these factors would fail to moti-vate non-borrowers to announce loans.We do?nd a signi?cantly positive coef?cient for the loan-to-assets ratio,suggesting that non-borrowers are more likely to announce loans that are material in this sense.These ?ndings con?rm our third hypothesis.

Our?ndings have implications for the samples typi-cally drawn in loan announcement studies.These samples do not appear representative of the market as a whole. Rather,they will be over-weighted toward smaller,

Table5

Logit model estimates of the probability a loan will be announced by the

borrowing?rm and/or other entities.

The sample consists of800loans randomly selected from a population

of20,140loans in the DealScan database over the period1987–2004.In

columns1and2,the dependent variable takes a value of one if the loan

was announced by the borrowing?rm and zero otherwise.The informa-

tion asymmetry index is the average quintile ranking of the borrowing

?rm based on six information asymmetry measures—forecast error,

forecast dispersion,bid-ask spread,?rm age,volatility of abnormal

returns around earnings announcement,and residual volatility of sock

returns.The dummy variable Negative EBITDA takes a value of one for

borrowers with zero or negative EBITDA and zero otherwise.Loan-to-

assets ratio is the log of one plus the loan-to-assets ratio.Market cap is

the log of market capitalization.Columns3and4show the estimates of

a multinomial logit model.The dependent variable takes unit value for

loans announced by the company and a value of two for loans

announced by other entities.The reference group is unannounced loans.

Column3shows estimates for announcements by the company and

column4for announcements by others.The numbers in parentheses are

standard errors.

nn Signi?cant at1%level;n5%level.

Dependent variable:

announced by

(1)(2)(3)(4)

Company Company Company Others

Interceptà3.3214nn 2.274à3.293nnà2.640nn

(0.368)(1.029)(0.373)(0.500)

Info.asym.index0.443nn0.432nnà0.051

(0.107)(0.109)(0.164)

Negative EBITDA 1.022nn0.920nn0.963nnà0.856

(0.239)(0.240)(0.243)(0.614)

Loan-to-asset ratio 1.515nn 1.372nn 2.059nn 2.650nn

(0.382)(0.385)(0.416)(0.523)

Market capà0.210nn

(0.051)

LR Chi-sq63.9nn64.1nn1018.9nn

N714714714

10Gomes and Phillips(2010)use?rm size as a measure of

information asymmetry and the correlation matrix in Table2shows

that market capitalization correlates best with the information asym-

metry index across the variables we employ.

11Our data suggest,however,that reporters and other entities tend

to disclose loans before the closing date,while borrowers typically

report on or after the close.In our sample,50%of announcements by

non-borrowers occurred before the loan closed,but only about10%of

the borrowers generated announcements prior to closing.Announce-

ments by others could preempt borrowers from announcing their loans.

Because the reference group in our multinomial model is unannounced

loans,the information asymmetry index could have a positive coef?cient

estimate.

P.K.Maskara,D.J.Mullineaux/Journal of Financial Economics101(2011)684–694691

cash-constrained ?rms posing serious information asym-metries because these borrowers are most likely to announce their loans.This prompts the following ques-tion:In the absence of systematic differences between borrowers with announced and unannounced loans,would loan announcements generate positive equity market response?

4.Event-study results with a representative sample We have shown that ?rms making bank loan announcements have well-de?ned characteristics that differentiate them from borrowing ?rms in general.If announcing ?rms fail to represent the universe of bor-rowers,what are the implications for the existing body of event-study research on loan announcements?We try to answer by examining whether the positive equity returns found in prior studies of bank loan announcements would survive if the samples had represented the borrower universe.We do so by computing a weighted-average abnormal return on a sample of loan announcements gathered by Billett,Flannery,and Gar?nkel (1995)(here-after BFG),where the weights re?ect the borrower uni-verse more accurately.We thank these authors for providing their data.

In Table 6,we array all the ?rms in the BFG sample into deciles by market capitalization,using the size breakpoints from Kenneth French’s data library.We also show in Table 6that the information asymmetry index for the ?rms in our randomly selected sample of 800loans decreases monotonically when arrayed by size in the same fashion.We calculate the mean abnormal returns for BFG’s loan announcements for each decile.Like Billett,Flannery,and Gar?nkel (1995),we calculate one-day market-adjusted abnormal returns,but the results do not change if we calculate two-day returns or use other speci?cations for predicted returns.We ?nd that 35%of

the loan borrowers in the sample are from the smallest decile,while only 2%fall in the largest decile.Thus,BFG’s sample is not representative of the population of loan borrowers as a whole.The abnormal return for the smallest ?rms is 148basis points.For the remainder of the sample,the announcement return is an insigni?cant 26basis points,and abnormal returns fail the signi?cance test in every decile.The statistically signi?cant return of about 68basis points observed by Billett,Flannery,and Gar?nkel (1995)clearly is driven by the ?rms in the smallest decile,and their sample is heavily weighted toward such ?rms.But what would we observe if bor-rowers were drawn from a sample that small ?rms do not dominate?To estimate such an outcome,we calculate the mean abnormal return weighted equally across each decile in the BFG sample.The result is 35basis points,which is statistically insigni?cant.

Perhaps small ?rms are not only more likely to announce loans but are also more likely to rely on them,as some studies imply (Maskara and Mullineaux,2011).If so,over-weighting small ?rms in loan announcement studies could be justi?ed.Our own random sample of 800loans is also skewed toward smaller ?rms (Table 6,column 7),but less so than BFG’s.And the results in Table 1con?rm that our sample conforms to the DealScan loan universe.So we compute the average abnormal return from BFG’s data using decile weights from our random sample of 800loans and the result is a statisti-cally insigni?cant 54basis points.12We also calculate the announcement return for our sample and get an abnormal return of a statistically insigni?cant 5basis points.We

Table 6

Abnormal returns to loan announcements across ?rm size deciles and weighted abnormal returns.

We calculate ?rm size deciles based on the Fama-French size breakpoints for all borrowing ?rms in the month of loan closing.Column 1shows the mean information asymmetry index value for ?rms in the appropriate size decile estimated based on randomly selected 800loans in our sample.Column 2shows the number of loans in the Billett,Flannery,and Gar?nkel (1995)sample in each size decile.Their sample includes 626‘‘clean’’loan announcements during the period 1980to 1989that are not contaminated by any confounding corporate events and have returns based on actual transaction prices.Column 3shows the mean one-day market-adjusted abnormal return on the loan announcement day for the BFG ?rms.Column 4shows standard deviations of the abnormal returns.Column 5shows the percent of BFG ?rms in each size decile experiencing positive announcement return.Column 6shows the percentage weight allotted to ?rms from each size decile in the BFG sample.Column 7shows the percentage of borrowing ?rms in each decile in the random sample of 800loans.nn

Signi?cant at 1%level.Mean info.asym.index

(1)

(2)(3)

(4)(5)

(6)

(7)

Firm decile N Abnormal return

Std dev Percent positive

Weight in BFG sample (%)

Weight in our sample (%)

4.03Smallest

2190.0148nn 0.0597*******.4821080.00450.0375*******.123710.00120.03475111103.074470.00220.021953892.815320.00300.021359582.716430.00060.025053752.717400.00340.021955662.338250.00150.025956482.389260.00140.016362472.12

Biggest 150.0023

0.0240472

10

All ?rms

626

0.0068nn

0.0428

53

12

We use bootstrapped standard errors to calculate the t -statistics.

We draw appropriate number of observations from each decile to re?ect the weight of our sample ?rms in each decile and calculate the mean.We perform 1,000iterations of this process to calculate the standard deviation of the mean abnormal return.

P.K.Maskara,D.J.Mullineaux /Journal of Financial Economics 101(2011)684–694

692

conclude that signi?cant loan announcement returns are unlikely outcomes when samples more accurately repre-sent the distribution of borrowers.We also infer that if all loans were to be announced,an event study on this universe would yield insigni?cant results.

Our?ndings suggest that bank loans,in general,are not special.Accordingly,the literature’s critical distinc-tions between private and public debt may be unjusti?ed. Although we?nd that loan announcements by the smal-lest BFG?rms result in signi?cantly positive abnormal returns,we refrain from drawing inferences about whether bank loans to very small borrowers are special. While we show that small borrowers are more likely to announce their loans,the majority of such borrowers (60%of such in our sample)apparently do not announce. In addition,the estimated return for borrowers in the smallest decile in our own sample is a statistically insig-ni?cant17basis points.

An alternative explanation,as the referee points out, for the results generated from the BFG sample could be that positive effects of loan announcements on equity returns for large?rms are too small to be detected. Though possible,it seems unlikely that we would detect responses for only the smallest category of?rms.If our observed results re?ect non-detection,we should observe monotonically decreasing abnormal returns across?rm size deciles.But the data in Table6show that the abnormal returns in the BFG sample drop sharply for ?rms in the second decile,and thereafter the abnormal returns follow no meaningful pattern.

5.Conclusion

Our results suggest that loan announcements are relatively rare events that emanate from speci?c types of?rms.Borrowers are more likely to announce loans when they present sizeable information asymmetries to investors or when the loans in question appear material. We treat loans as material when they are a large compo-nent of the borrower’s capital structure or when the borrower faces higher prospects of cash?ow problems.

Firms that announce loans do not represent the popu-lation of borrowers,and this raises questions about whether the loan announcement results reported in the literature apply broadly.We?nd that the signi?cant abnormal returns in BFG’s loan sample are limited to the smallest10%of the borrower universe.If we apply weights to the BFG results that are more representative of borrowing?rms,in general,then the overall estimated abnormal return is insigni?cantly different from zero.The estimated abnormal return for the?rms in our own sample of announcements is likewise insigni?cantly dif-ferent from zero.Given these results,we infer that if all loans were to be announced,an event study on such a sample would fail to produce signi?cant positive returns.

Do these results in?uence the literature’s consensus that loans represent‘‘special’’types of?nancings and that public and private debt differ in critical ways?We con-tend that they do.Although loan announcements might elicit positive abnormal returns in special cases,the conclusion that loans are unique does not generalize to the population of loans.Public and private debt conse-quently could be cut from the same rather than a different cloth.

Our research focuses on the prospect of self-selection in disclosure decisions in a speci?c context.Yet?rms can choose to announce any of a broad array of outcomes not subject to regulatory disclosures.For example,disclosure of operational events like obtaining a large sales contract or forming strategic,marketing,or technological partner-ships is discretionary,but small?rms facing higher information asymmetries may be more likely to make such announcements.Consequently,the issue we raise could be relevant to event-study research more broadly. References

Best,R.,Zhang,H.,1993.Alternative information sources and the information content of bank loans.The Journal of Finance48, 1507–1522.

Bharath,S.T.,Pasquariello,P.,Wu,G.,2009.Does information asymmetry drive capital structure decisions?Review of Financial Studies22, 3211–3243

Billett,M.T.,Flannery,M.J.,Gar?nkel,J.A.,1995.The effect of lender identity on a borrowing?rm’s equity return.The Journal of Finance 50,699–718.

Billett,M.T.,Flannery,M.J.,Gar?nkel,J.A.,2006.Are bank loans special?

Evidence on the post-announcement performance of bank bor-rowers.Journal of Financial and Quantitative Analysis41,733–751. Chaney,P.K.,Jeter,D.C.,Shivakumar,L.,2004.Self-selection of auditors and audit pricing in private?rms.The Accounting Review79,51–72. Chung,K.,McInish,T.,Wood,R.,Wyhowski, D.,1995.Production of information,information asymmetry,and the bid-ask spread.Jour-nal of Banking and Finance19,1025–1046.

Chung,K.,Zhang,H.,2009.A Simple Approximation of Intraday Spreads Using Daily Data.Unpublished Working Paper.State University of New York,Buffalo.

Clarke,J.,Shastri,K.,2001.On Information Asymmetry Metrics.Unpub-lished Working Paper.University of Pittsburgh.

Corwin,S.,Schultz,P.,2010.A Simple Way to Estimate Bid-ask Spreads from Daily High and Low Prices.Unpublished Working Paper.

University of Notre Dame.

Diamond,D.,1985.Optimal release of information by?rms.Journal of Finance40,1071–1094.

Diamond,D.,Verrecchia,R.,1991.Disclosure,liquidity,and the cost of capital.Journal of Finance46,1325–1359.

Dierkens,N.,https://www.sodocs.net/doc/6e880725.html,rmation asymmetry and equity issues.Journal of Financial and Quantitative Analysis26,181–199.

Easley,D.,Hvidkjaer,S.,O’Hara,M.,2010.Factoring information into returns.Journal of Financial and Quantitative Analysis45,293–309. Fama, E.,1985.What’s different about banks?Journal of Monetary Economics15,29–39

Focarelli, D.,Pozzolo, A.F.,Casolaro,L.,2008.The pricing effect of certi?cation on syndicated loans.Journal of Monetary Economics55, 335–349.

Gande,A.,Saunders,A.,2006.Are Banks Still Special when there is a Secondary Market for Loans?Unpublished Working Paper.Southern Methodist University and New York University.

Gomes,A.,Phillips,G.,2010.Private and Public Security Issuance by Public Firms:The Role of Asymmetric Information.Unpublished Working Paper.Washington University.

James,C.,1987.Some evidence on the uniqueness of bank loans.Journal of Financial Economics19,217–236.

Krishnaswami,S.,Spindt,P.,Subramaniam,V.,https://www.sodocs.net/doc/6e880725.html,rmation asym-metry,monitoring,and the placement structure of corporate debt.

Journal of Financial Economics51,407–434.

Krishnaswami,S.,Subramaniam,V.,https://www.sodocs.net/doc/6e880725.html,rmation asymmetry, valuation,and the corporate spin-off decision.Journal of Financial Economics53,73–112.

Leary,M.T.,Roberts,M.R.,2010.The pecking order,debt capacity,and information asymmetry.Journal of Financial Economics95,332–355. Li,K.,Prabhala,N.,2007.Self-selection models in corporate?nance.In: Eckbo,E.(Ed.),A Handbook of Corporate Finance,vol.1:Empirical Corporate Finance.Elsevier,North-Holland,pp.39–85.

P.K.Maskara,D.J.Mullineaux/Journal of Financial Economics101(2011)684–694693

Lummer,S.,McConnell,J.,1989.Further evidence on the bank lending process and the capital market response to bank loan agreements.

Journal of Financial Economics25,99–122.

Maskara,P.,Mullineaux,D.J.,2011.Small?rm capital structure and the syndicated loan market.Journal of Financial Services Research39,55–70. Marsh,I.W.,2006.The Effect of Lender’s Credit Risk Transfer Activities on Borrowing Firms’Equity Returns.Unpublished Working Paper.

City University London.

McNichols,M.,O’Brien,P.,1997.Self-selection and analyst coverage.

Journal of Accounting Research35,167–199.

Mosebach,M.,1999.Market responses to banks granting lines of credit.

Journal of Banking and Finance23,1707–1723.Preece,D.,Mullineaux,D.J.,1994.Monitoring by?nancial intermedi-aries:banks vs.nonbanks.Journal of Financial Services Research4, 191–200.

Preece,D.,Mullineaux,D.J.,1996.Monitoring,loan renegotiability,and ?rm value:the role of lending syndicates.Journal of Banking and Finance20,577–593.

Slovin,M.B.,Johnson,S.A.,Glascock,J.L.,1992.Firm size and the information content of bank loan announcements.Journal of Bank-ing and Finance16,1057–1071.

Verrecchia,R.,2001.Essays on disclosure.Journal of Accounting and Economics32,97–180.

P.K.Maskara,D.J.Mullineaux/Journal of Financial Economics101(2011)684–694 694

交互式多模型算法仿真与分析

硕037班 刘文3110038020 2011/4/20交互式多模型仿真与分析IMM算法与GBP算法的比较,算法实现和运动目标跟踪仿真,IMM算法的特性分析 多源信息融合实验报告

交互式多模型仿真与分析 一、 算法综述 由于混合系统的结构是未知的或者随机突变的,在估计系统状态参数的同时还需要对系统的运动模式进行辨识,所以不可能通过建立起一个固定的模型对系统状态进行效果较好的估计。针对这一问题,多模型的估计方法提出通过一个模型集{}(),1,2,,j M m j r == 中不同模型的切换来匹配不同目标的运动或者同一目标不同阶段的运动,达到运动模式的实时辨识的效果。 目前主要的多模型方法包括一阶广义贝叶斯方法(BGP1),二阶广义贝叶斯方法(GPB2)以及交互式多模型方法等(IMM )。这些多模型方法的共同点是基于马尔科夫链对各自的模型集进行切换或者融合,他们的主要设计流程如下图: M={m1,m2,...mk} K 时刻输入 值的形式 图一 多模型设计方法 其中,滤波器的重初始化方式是区分不同多模型算法的主要标准。由于多模型方法都是基于一个马尔科夫链来切换与模型的,对于元素为r 的模型集{}(),1,2,,j M m j r == ,从0时刻到k 时刻,其可能的模型切换轨迹为 120,12{,,}k i i i k trace k M m m m = ,其中k i k m 表示K-1到K 时刻,模型切换到第k i 个, k i 可取1,2,,r ,即0,k trace M 总共有k r 种可能。再令1 2 1 ,,,,k k i i i i μ+ 为K+1时刻经由轨迹0,k trace M 输入到第1k i +个模型滤波器的加权系数,则输入可以表示为 0,11 2 1 12|,,,,|,,,???k k trace k k k i M k k i i i i k k i i i x x μ++=?∑ 可见轨迹0,k trace M 的复杂度直接影响到算法计算量和存储量。虽然全轨迹的

五种大数据压缩算法

?哈弗曼编码 A method for the construction of minimum-re-dundancy codes, 耿国华1数据结构1北京:高等教育出版社,2005:182—190 严蔚敏,吴伟民.数据结构(C语言版)[M].北京:清华大学出版社,1997. 冯桂,林其伟,陈东华.信息论与编码技术[M].北京:清华大学出版社,2007. 刘大有,唐海鹰,孙舒杨,等.数据结构[M].北京:高等教育出版社,2001 ?压缩实现 速度要求 为了让它(huffman.cpp)快速运行,同时不使用任何动态库,比如STL或者MFC。它压缩1M数据少于100ms(P3处理器,主频1G)。 压缩过程 压缩代码非常简单,首先用ASCII值初始化511个哈夫曼节点: CHuffmanNode nodes[511]; for(int nCount = 0; nCount < 256; nCount++) nodes[nCount].byAscii = nCount; 其次,计算在输入缓冲区数据中,每个ASCII码出现的频率: for(nCount = 0; nCount < nSrcLen; nCount++) nodes[pSrc[nCount]].nFrequency++; 然后,根据频率进行排序: qsort(nodes, 256, sizeof(CHuffmanNode), frequencyCompare); 哈夫曼树,获取每个ASCII码对应的位序列: int nNodeCount = GetHuffmanTree(nodes); 构造哈夫曼树 构造哈夫曼树非常简单,将所有的节点放到一个队列中,用一个节点替换两个频率最低的节点,新节点的频率就是这两个节点的频率之和。这样,新节点就是两个被替换节点的父

LZ77压缩算法实验报告

LZ77压缩算法实验报告 一、实验内容 使用C++编程实现LZ77压缩算法的实现。 二、实验目的 用LZ77实现文件的压缩。 三、实验环境 1、软件环境:Visual C++ 6.0 2、编程语言:C++ 四、实验原理 LZ77 算法在某种意义上又可以称为“滑动窗口压缩”,这是由于该算法将一个虚拟的,可以跟随压缩进程滑动的窗口作为术语字典,要压缩的字符串如果在该窗口中出现,则输出其出现位置和长度。使用固定大小窗口进行术语匹配,而不是在所有已经编码的信息中匹配,是因为匹配算法的时间消耗往往很多,必须限制字典的大小才能保证算法的效率;随着压缩的进程滑动字典窗口,使其中总包含最近编码过的信息,是因为对大多数信息而言,要编码的字符串往往在最近的上下文中更容易找到匹配串。 五、LZ77算法的基本流程 1、从当前压缩位置开始,考察未编码的数据,并试图在滑动窗口中找出最长的匹 配字符串,如果找到,则进行步骤2,否则进行步骤3。 2、输出三元符号组( off, len, c )。其中off 为窗口中匹

配字符串相对窗口边 界的偏移,len 为可匹配的长度,c 为下一个字符。然后将窗口向后滑动len + 1 个字符,继续步骤1。 3、输出三元符号组( 0, 0, c )。其中c 为下一个字符。然后将窗口向后滑动 len + 1 个字符,继续步骤1。 六、源程序 /********************************************************************* * * Project description: * Lz77 compression/decompression algorithm. * *********************************************************************/ #include #include #include #include #define OFFSET_CODING_LENGTH (10) #define MAX_WND_SIZE 1024 //#define MAX_WND_SIZE (1<

LZSS压缩算法实验报告

实验名称:LZSS压缩算法实验报告 一、实验内容 使用Visual 6..0 C++编程实现LZ77压缩算法。 二、实验目的 用LZSS实现文件的压缩。 三、实验原理 LZSS压缩算法是词典编码无损压缩技术的一种。LZSS压缩算法的字典模型使用了自适应的方式,也就是说,将已经编码过的信息作为字典, 四、实验环境 1、软件环境:Visual C++ 6.0 2、编程语言:C++ 五、实验代码 #include #include #include #include /* size of ring buffer */ #define N 4096 /* index for root of binary search trees */ #define NIL N /* upper limit for g_match_len. Changed from 18 to 16 for binary compatability with Microsoft COMPRESS.EXE and EXPAND.EXE #define F 18 */ #define F 16 /* encode string into position and length if match_length is greater than this: */ #define THRESHOLD 2 /* these assume little-endian CPU like Intel x86

-- need byte-swap function for big endian CPU */ #define READ_LE32(X) *(uint32_t *)(X) #define WRITE_LE32(X,Y) *(uint32_t *)(X) = (Y) /* this assumes sizeof(long)==4 */ typedef unsigned long uint32_t; /* text (input) size counter, code (output) size counter, and counter for reporting progress every 1K bytes */ static unsigned long g_text_size, g_code_size, g_print_count; /* ring buffer of size N, with extra F-1 bytes to facilitate string comparison */ static unsigned char g_ring_buffer[N + F - 1]; /* position and length of longest match; set by insert_node() */ static unsigned g_match_pos, g_match_len; /* left & right children & parent -- these constitute binary search tree */ static unsigned g_left_child[N + 1], g_right_child[N + 257], g_parent[N + 1]; /* input & output files */ static FILE *g_infile, *g_outfile; /***************************************************************************** initialize trees *****************************************************************************/ static void init_tree(void) { unsigned i; /* For i = 0 to N - 1, g_right_child[i] and g_left_child[i] will be the right and left children of node i. These nodes need not be initialized. Also, g_parent[i] is the parent of node i. These are initialized to NIL (= N), which stands for 'not used.' For i = 0 to 255, g_right_child[N + i + 1] is the root of the tree for strings that begin with character i. These are initialized to NIL. Note there are 256 trees. */ for(i = N + 1; i <= N + 256; i++) g_right_child[i] = NIL; for(i = 0; i < N; i++) g_parent[i] = NIL; } /***************************************************************************** Inserts string of length F, g_ring_buffer[r..r+F-1], into one of the trees (g_ring_buffer[r]'th tree) and returns the longest-match position and length via the global variables g_match_pos and g_match_len. If g_match_len = F, then removes the old node in favor of the new one, because the old one will be deleted sooner.

多媒体数据压缩实验报告

多媒体数据压缩实验报告 篇一:多媒体实验报告_文件压缩 课程设计报告 实验题目:文件压缩程序 姓名:指导教师:学院:计算机学院专业:计算机科学与技术学号: 提交报告时间:20年月日 四川大学 一,需求分析: 有两种形式的重复存在于计算机数据中,文件压缩程序就是对这两种重复进行了压 缩。 一种是短语形式的重复,即三个字节以上的重复,对于这种重复,压缩程序用两个数字:1.重复位置距当前压缩位置的距离;2.重复的长度,来表示这个重复,假设这两个数字各占一个字节,于是数据便得到了压缩。 第二种重复为单字节的重复,一个字节只有256种可能的取值,所以这种重复是必然的。给 256 种字节取值重新编码,使出现较多的字节使用较短的编码,出现较少的字节使用较长的编码,这样一来,变短的字节相对于变长的字节更多,文件的总长度就会减少,并且,字节使用比例越不均

匀,压缩比例就越大。 编码式压缩必须在短语式压缩之后进行,因为编码式压缩后,原先八位二进制值的字节就被破坏了,这样文件中短语式重复的倾向也会被破坏(除非先进行解码)。另外,短语式压缩后的结果:那些剩下的未被匹配的单、双字节和得到匹配的距离、长度值仍然具有取值分布不均匀性,因此,两种压缩方式的顺序不能变。 本程序设计只做了编码式压缩,采用Huffman编码进行压缩和解压缩。Huffman编码是一种可变长编码方式,是二叉树的一种特殊转化形式。编码的原理是:将使用次数多的代码转换成长度较短的代码,而使用次数少的可以使用较长的编码,并且保持编码的唯一可解性。根据 ascii 码文件中各 ascii 字符出现的频率情况创建 Huffman 树,再将各字符对应的哈夫曼编码写入文件中。同时,亦可根据对应的哈夫曼树,将哈夫曼编码文件解压成字符文件. 一、概要设计: 压缩过程的实现: 压缩过程的流程是清晰而简单的: 1. 创建 Huffman 树 2. 打开需压缩文件 3. 将需压缩文件中的每个 ascii 码对应的 huffman 编码按 bit 单位输出生成压缩文件压缩结束。

数据快速压缩算法的C语言实现

价值工程 置,是一项十分有意义的工作。另外恶意代码的检测和分析是一个长期的过程,应对其新的特征和发展趋势作进一步研究,建立完善的分析库。 参考文献: [1]CNCERT/CC.https://www.sodocs.net/doc/6e880725.html,/publish/main/46/index.html. [2]LO R,LEVITTK,OL SSONN R.MFC:a malicious code filter [J].Computer and Security,1995,14(6):541-566. [3]KA SP ER SKY L.The evolution of technologies used to detect malicious code [M].Moscow:Kaspersky Lap,2007. [4]LC Briand,J Feng,Y Labiche.Experimenting with Genetic Algorithms and Coupling Measures to devise optimal integration test orders.Software Engineering with Computational Intelligence,Kluwer,2003. [5]Steven A.Hofmeyr,Stephanie Forrest,Anil Somayaji.Intrusion Detection using Sequences of System calls.Journal of Computer Security Vol,Jun.1998. [6]李华,刘智,覃征,张小松.基于行为分析和特征码的恶意代码检测技术[J].计算机应用研究,2011,28(3):1127-1129. [7]刘威,刘鑫,杜振华.2010年我国恶意代码新特点的研究.第26次全国计算机安全学术交流会论文集,2011,(09). [8]IDIKA N,MATHUR A P.A Survey of Malware Detection Techniques [R].Tehnical Report,Department of Computer Science,Purdue University,2007. 0引言 现有的压缩算法有很多种,但是都存在一定的局限性,比如:LZw [1]。主要是针对数据量较大的图像之类的进行压缩,不适合对简单报文的压缩。比如说,传输中有长度限制的数据,而实际传输的数据大于限制传输的数据长度,总体数据长度在100字节左右,此时使用一些流行算法反而达不到压缩的目的,甚至增大数据的长度。本文假设该批数据为纯数字数据,实现压缩并解压缩算法。 1数据压缩概念 数据压缩是指在不丢失信息的前提下,缩减数据量以减少存储空间,提高其传输、存储和处理效率的一种技术方法。或按照一定的算法对数据进行重新组织,减少数据的冗余和存储的空间。常用的压缩方式[2,3]有统计编码、预测编码、变换编码和混合编码等。统计编码包含哈夫曼编码、算术编码、游程编码、字典编码等。 2常见几种压缩算法的比较2.1霍夫曼编码压缩[4]:也是一种常用的压缩方法。其基本原理是频繁使用的数据用较短的代码代替,很少使用 的数据用较长的代码代替,每个数据的代码各不相同。这些代码都是二进制码,且码的长度是可变的。 2.2LZW 压缩方法[5,6]:LZW 压缩技术比其它大多数压缩技术都复杂,压缩效率也较高。其基本原理是把每一个第一次出现的字符串用一个数值来编码,在还原程序中再将这个数值还成原来的字符串,如用数值0x100代替字符串ccddeee"这样每当出现该字符串时,都用0x100代替,起到了压缩的作用。 3简单报文数据压缩算法及实现 3.1算法的基本思想数字0-9在内存中占用的位最 大为4bit , 而一个字节有8个bit ,显然一个字节至少可以保存两个数字,而一个字符型的数字在内存中是占用一个字节的,那么就可以实现2:1的压缩,压缩算法有几种,比如,一个自己的高四位保存一个数字,低四位保存另外一个数字,或者,一组数字字符可以转换为一个n 字节的数值。N 为C 语言某种数值类型的所占的字节长度,本文讨论后一种算法的实现。 3.2算法步骤 ①确定一种C 语言的数值类型。 —————————————————————— —作者简介:安建梅(1981-),女,山西忻州人,助理实验室,研究方 向为软件开发与软交换技术;季松华(1978-),男,江苏 南通人,高级软件工程师,研究方向为软件开发。 数据快速压缩算法的研究以及C 语言实现 The Study of Data Compression and Encryption Algorithm and Realization with C Language 安建梅①AN Jian-mei ;季松华②JI Song-hua (①重庆文理学院软件工程学院,永川402160;②中信网络科技股份有限公司,重庆400000)(①The Software Engineering Institute of Chongqing University of Arts and Sciences ,Chongqing 402160,China ; ②CITIC Application Service Provider Co.,Ltd.,Chongqing 400000,China ) 摘要:压缩算法有很多种,但是对需要压缩到一定长度的简单的报文进行处理时,现有的算法不仅达不到目的,并且变得复杂, 本文针对目前一些企业的需要,实现了对简单报文的压缩加密,此算法不仅可以快速对几十上百位的数据进行压缩,而且通过不断 的优化,解决了由于各种情况引发的解密错误,在解密的过程中不会出现任何差错。 Abstract:Although,there are many kinds of compression algorithm,the need for encryption and compression of a length of a simple message processing,the existing algorithm is not only counterproductive,but also complicated.To some enterprises need,this paper realizes the simple message of compression and encryption.This algorithm can not only fast for tens of hundreds of data compression,but also,solve the various conditions triggered by decryption errors through continuous optimization;therefore,the decryption process does not appear in any error. 关键词:压缩;解压缩;数字字符;简单报文Key words:compression ;decompression ;encryption ;message 中图分类号:TP39文献标识码:A 文章编号:1006-4311(2012)35-0192-02 ·192·

压缩编码算法设计与实现实验报告

压缩编码算法设计与实现实验报告 一、实验目的:用C语言或C++编写一个实现Huffman编码的程序,理解对数据进行无损压缩的原理。 二、实验内容:设计一个有n个叶节点的huffman树,从终端读入n个叶节 点的权值。设计出huffman编码的压缩算法,完成对n个节点的编码,并写出程序予以实现。 三、实验步骤及原理: 1、原理:Huffman算法的描述 (1)初始化,根据符号权重的大小按由大到小顺序对符号进行排序。 (2)把权重最小的两个符号组成一个节点, (3)重复步骤2,得到节点P2,P3,P4……,形成一棵树。 (4)从根节点开始顺着树枝到每个叶子分别写出每个符号的代码 2、实验步骤: 根据算法原理,设计程序流程,完成代码设计。 首先,用户先输入一个数n,以实现n个叶节点的Huffman 树; 之后,输入n个权值w[1]~w[n],注意是unsigned int型数值; 然后,建立Huffman 树; 最后,完成Huffman编码。 四、实验代码:#include #include #include #define MAX 0xFFFFFFFF typedef struct / /*设计一个结构体,存放权值,左右孩子*// { unsigned int weight; unsigned int parent,lchild,rchild; }HTNode,*HuffmanTree; typedef char** HuffmanCode;

int min(HuffmanTree t,int i) { int j,flag; unsigned int k = MAX; for(j=1;j<=i;j++) if(t[j].parent==0&&t[j].weight s2) { tmp = s1; s1 = s2; s2 = tmp; } } void HuffmanCoding(HuffmanTree &HT,HuffmanCode &HC,int *w,int n,int &wpl) { int m,i,s1,s2,start,j; unsigned int c,f; HuffmanTree p; char *cd; if(n<=1) return; m=2*n-1; HT=(HuffmanTree)malloc((m+1)*sizeof(HTNode)); for(p=HT+1,i=1;i<=n;++i,++p,++w) { (*p).weight=*w; (*p).parent=0; (*p).lchild=0; (*p).rchild=0; }

交互式多模型算法卡尔曼滤波仿真

交互式多模型算法卡尔曼滤波仿真 1 模型建立 分别以加速度u=0、1、2代表三个不同的运动模型 状态方程x(k+1)=a*x(k)+b*w(k)+d*u 观察方程z(k)=c*x(k)+v(k) 其中,a=[1 dt;0 1],b=[dt^2/2;dt],d=[dt^2/2;dt],c=[1 0] 2 程序代码 由两个功能函数组成,imm_main用来实现imm 算法,move_model1用来生成仿真数据,初始化运动参数 function imm_main %交互式多模型算法主程序 %初始化有关参数 move_model %调用运动模型初始化及仿真运动状态生成函数 load movedata %调入有关参数初始值(a b d c u position velocity pmeas dt tg q r x_hat p_var) p_tran=[0.8 0.1 0.1;0.2 0.7 0.1;0.1 0.2 0.7];%转移概率 p_pri=[0.1;0.6;0.3];%模型先验概率 n=1:2:5; %提取各模型方差矩阵 k=0; %记录仿真步数 like=[0;0;0];%视然函数 x_hat_hat=zeros(2,3);%三模型运动状态重初始化矩阵 u_=zeros(3,3);%混合概率矩阵 c_=[0;0;0];%三模型概率更新系数 %数据保存有关参数初始化 phat=[];%保存位置估值 vhat=[];%保存速度估值 xhat=[0;0];%融合和运动状态 z=0;%量测偏差(一维位置) pvar=zeros(2,2);%融合后估计方差 for t=0:dt:tg; %dt为为仿真步长;tg为仿真时间长度 k=k+1;%记录仿真步数 ct=0; %三模型概率更新系数 c_max=[0 0 0];%混合概率规范系数 p_var_hat=zeros(2,6);%方差估计重初始化矩阵, %[x_hat_hat p_var_hat]=model_reinitialization(p_tran,p_pri,x_hat,p_var);%调用重初始化函数,进行混合估计,生成三个模型重初始化后的运动状态、方差 %混合概率计算 for i=1:3 u_(i,:)=p_tran(i,:)*p_pri(i); end for i=1:3 c_max=c_max+u_(i,:); end

数据压缩实验指导书

目 录 实验一用C/C++语言实现游程编码 实验二用C/C++语言实现算术编码 实验三用C/C++语言实现LZW编码 实验四用C/C++语言实现2D-DCT变换13

实验一用C/C++语言实现游程编码 1. 实验目的 1) 通过实验进一步掌握游程编码的原理; 2) 用C/C++语言实现游程编码。 2. 实验要求 给出数字字符,能正确输出编码。 3. 实验内容 现实中有许多这样的图像,在一幅图像中具有许多颜色相同的图块。在这些图块中,许多行上都具有相同的颜色,或者在一行上有许多连续的象素都具有相同的颜色值。在这种情况下就不需要存储每一个象素的颜色值,而仅仅存储一个象素的颜色值,以及具有相同颜色的象素数目就可以,或者存储一个象素的颜色值,以及具有相同颜色值的行数。这种压缩编码称为游程编码,常用(run length encoding,RLE)表示,具有相同颜色并且是连续的象素数目称为游程长度。 为了叙述方便,假定一幅灰度图像,第n行的象素值为: 用RLE编码方法得到的代码为:0@81@38@501@40@8。代码中用黑体表示的数字是游程长度,黑体字后面的数字代表象素的颜色值。例如黑体字50代表有连续50个象素具有相同的颜色值,它的颜色值是8。 对比RLE编码前后的代码数可以发现,在编码前要用73个代码表示这一行的数据,而编码后只要用11个代码表示代表原来的73个代码,压缩前后的数据量之比约为7:1,即压缩比为7:1。这说明RLE确实是一种压缩技术,而且这种编码技术相当直观,也非常经济。RLE所能获得的压缩比有多大,这主要是取决于图像本身的特点。如果图像中具有相同颜色的图像块越大,图像块数目越少,获得的压缩比就越高。反之,压缩比就越小。 译码时按照与编码时采用的相同规则进行,还原后得到的数据与压缩前的数据完全相同。因此,RLE是无损压缩技术。

数据压缩实验

实验二图像预测编码 一、实验题目: 图像预测编码: 二、实验目的: 实现图像预测编码和解码. 三、实验内容: 给定一幅图片,对其进行预测编码,获得预测图像,绝对残差图像, 再利用预测图像和残差图像进行原图像重建并计算原图像和重建图像误差. 四、预备知识: 预测方法,图像处理概论。 五、实验原理: 根据图像中任意像素与周围邻域像素之间存在紧密关系,利用周围邻域四个像素来进行该点像素值预测,然后传输图像像素值与其预测值的差值信号,使传输的码率降低,达到压缩的目的。 六、实验步骤: (1)选取一幅预测编码图片; (2)读取图片内容像素值并存储于矩阵; (3)对图像像素进行预测编码; (4)输出预测图像和残差图像; (5)根据预测图像和残差图像重建图像; (6)计算原预测编码图像和重建图像误差. 七、思考题目: 如何采用预测编码算法实现彩色图像编码和解码. 八、实验程序代码: 预测编码程序1: 编码程序: i1=imread('lena.jpg'); if isrgb(i1) i1=rgb2gray(i1);

end i1=imcrop(i1,[1 1 256 256]); i=double(i1); [m,n]=size(i); p=zeros(m,n); y=zeros(m,n); y(1:m,1)=i(1:m,1); p(1:m,1)=i(1:m,1); y(1,1:n)=i(1,1:n); p(1,1:n)=i(1,1:n); y(1:m,n)=i(1:m,n); p(1:m,n)=i(1:m,n); p(m,1:n)=i(m,1:n); y(m,1:n)=i(m,1:n); for k=2:m-1 for l=2:n-1 y(k,l)=(i(k,l-1)/2+i(k-1,l)/4+i(k-1,l-1)/8+i(k-1,l+1)/8); p(k,l)=round(i(k,l)-y(k,l)); end end p=round(p); subplot(3,2,1); imshow(i1); title('原灰度图像'); subplot(3,2,2); imshow(y,[0 256]); title('利用三个相邻块线性预测后的图像'); subplot(3,2,3); imshow(abs(p),[0 1]); title('编码的绝对残差图像'); 解码程序 j=zeros(m,n); j(1:m,1)=y(1:m,1); j(1,1:n)=y(1,1:n); j(1:m,n)=y(1:m,n);

LZ77 压缩算法实验报告 一

LZ77 压缩算法实验报告 一、实验内容:使用 C++编程实现 LZ77 压缩算法的实现。 二、实验目的:用 LZ77 实现文件的压缩。 三、实验环境: 1、软件环境:Visual C++ 6.0 2、编程语言:C++ 四、实验原理: LZ77 算法在某种意义上又可以称为“滑动窗口压缩”,这是 由于该算法将一个虚拟的,可以跟随压缩进程滑动的窗口作为术语字典,要压缩的字符串如果在该窗口中出现,则输出其出现位置和长度。使用固定大小窗口进行术语匹配,而不是在所有已经编码的信息中匹配,是因为匹配算法的时间消耗往往很多,必须限制字典的大小才能保证算法的效率;随着压缩的进程滑动字典窗口,使其中总包含最近编码过的信息,是因为对大多数信息而言,要编码的字符串往往在最近的上下文中更容易找到匹配串。 五、 LZ77 算法的基本流程:1、从当前压缩位置开始,考察未编码的数据,并 试图在滑动窗口中找出最长的匹配字符串,如果找到,则进行步骤2,否则进行步骤 3。 2、输出三元符号组 ( off, len, c )。其中 off 为窗口中匹配字符串相 对窗口边界的偏移,len 为可匹配的长度,c 为下一个字符。然后将窗口向后滑动 len + 1 个字符,继续步骤 1。 3、输出三元符号组 ( 0, 0, c )。其中 c 为下一个字符。然后将窗口向 后滑动 len + 1 个字符,继续步骤 1。 代码如下: #include #include #include #include"lz77.h" ////////////////////////////////////////////////////////////////// // out file format: // 0;flag2;buffer;0;flag2;buffer;...flag1;flag2;bufferlast // flag1 - 2 bytes, source buffer block length // if block size is 65536, be zero // flag2 - 2 bytes, compressed buffer length // if can not compress, be same with flag1 ////////////////////////////////////////////////////////////////// void main(int argc, char* argv[]) { /*if (argc != 4) { puts("Usage: ");

哈夫曼算法实现字符串压缩——实验报告单

《用哈夫曼编码实现文件压缩》 实验报告 课程名称《数据结构B》 实验学期 2011 至 2012 学年第一学期学生所在系部计算机系 年级 2009级专业班级计科B09—1 学生姓名韩翼学号 200907014106 任课教师盛建瓴 实验成绩

一、实验题目: 用哈夫曼编码实现文件压缩 二、实验目的: 1、了解文件的概念。 2、掌握线性链表的插入、删除等算法。 3、掌握Huffman 树的概念及构造方法。 4、掌握二叉树的存储结构及遍历算法。 5、利用Huffman 树及Huffman 编码,掌握实现文件压缩的一般原理。 三、实验设备与环境: 微型计算机、Windows 系列操作系统 、Visual C++6.0软件 四、实验内容: 根据ascii 码文件中各ascii 字符出现的频率情况创建Haffman 树,再将各字符对应的哈夫曼编码写入文件中,实现文件压缩。 五、概要设计: 本次试验采用将字符用长度尽可能短的二进制数位表示方法,即对于文件中出现的字符,无须全部都用8位的ASCLL 码进行存储,根据他们在文件中出现的频率不同,我们利用Haffman 算法使每个字符能以最短的二进制字符进行存储,以达到节省存储空间,压缩文件的目的。解决了压缩需采用的算法,程序的思路已然清晰: 1、统计需压缩文件中每个字符出现的频率。 2、将每个字符的出现频率作为叶子结点构建Haffman 树,然后将树中结点 引向其左孩子的分支标“0”,引向其右孩子的分支标“1” ; 每个字符的编码即为从根到每个叶子的路径上得到的0、1序列,这样便完成了Haffman 编码,将每个字符用最短的二进制字符表示。 3、打开需压缩的文件,再将需压缩文件中的每个ASCII 码对应的编码按bit 单位输出。 4、文件压缩结束。 六、详细设计: (1)Huffman 树简介 路径:从树中一个结点到另一个结点之间的分支构成这两个结点间的路径 路径长度:路径上的分支数 树的路径长度:从树根到每一个结点的路径长度之和 树的带权路径长度:树中所有带权结点的路径长度之和 —结点到根的路径长度 ——权值 —其中:记作:k k n k k k l w l w wpl ∑==1

《数据压缩与信源编码》实验指导书

《数据压缩与信源编码》实验指导书 适用专业:信息工程 课程代码: 6088619 总学时: 40 总学分: 2.5 编写单位:电气与电子信息学院 编写人:李斌 审核人: 审批人: 批准时间: 2015 年 11 月 10日

目录 实验一码书的设计和使用 (2) 实验二基于DCT变换的图像压缩技术 (8) 实验三基于小波变换的图像压缩技术 (15)

实验一 码书的设计和使用 一、实验目的 采用矢量量化算法(LBG )获得图像压缩所需要的码书,通过码书实现图像压缩编码。 二、实验内容 对给定的一幅图像进行码书设计、编码和解码。 三、实验仪器、设备及材料 操作系统:Windowsxp ; 软件:MA TLAB 四、实验原理 要想得到好的性能编码,仅采用标量量化是不可能的。当把多个信源符号联合起来形成多维矢量,再对矢量进行标量量化时自由度将更大,同样的失真下,量化基数可进一步减少,码率可进一步压缩。这种量化叫矢量量化。 一种有效和直观的矢量量化码书设计算法——LBG 算法(也叫GLA 算法)是由Linde 、Buzo 和Gray 于1980年首先提出来的。该算法基于最佳矢量量化器设计的最佳划分和最佳码书这两个必要条件,且是Lloyd 算法在矢量空间的推广,其特点为物理概念清晰、算法理论严密及算法实现容易。 设训练矢量集为{}110,,,-=M x x x X ,待产生的码书为{}110,,,-=N y y y C ,其中{})1(10,,,-=k i i i i x x x x ,{})1(10,,,-=k j j j j y y y y ,10,10-≤≤-≤≤N j M i ,则码书设计过程就是需求把训练矢量集X 分成N 个子集)1,,1,0(-=N j S j 的一种最佳聚类方案,而子集j S 的质心矢量j y 作为码字。假设平方误差测度用来表征训练矢量i x 和码字j y 之间的失真,即: ∑-=-=1 02)(),(k l jl il j i y x y x d 则码书设计的准则可用下列数学形式表达: 最小化 ∑∑-=-==101 0),(),,(N j M i j i ij y x d w C X W f 约束条件 ∑-==101N j ij w ,10-≤≤M i 其中W 为N M ?矩阵,其元素满足:

文本压缩算法的比较研究

第22卷 第12期 2006年12月 甘肃科技 Gansu Science and Technology V ol.22 N o.12Dec. 2006 文本压缩算法的比较研究 徐成俊1,舒 毅1,柴 蓉2,张其斌1,田全红1,郝 涛3 (1甘肃省计算中心,甘肃兰州730030;2西北师范大学,甘肃兰州730070;3兰州理工大学甘肃兰州730050) 摘 要:论述了4种不同的文本压缩算法。根据压缩算法的优点和缺点,在实践中,要有针对性选 择算法,用其优点,从而得到比较理想的压缩文本。关键词:压缩算法;哈夫曼;算术;L Z ;游程中图分类号:TP391 1 引言 随着计算机技术的快速发展,各种系统数据量越来越大,给信息存储特别是网络传输带来诸多的困难,己成为有效获取和使用信息的瓶颈。为了节省信息的存储空间和提高信息的传输效率,必须对大量的实际数据进行压缩。实践证明,采用数据压缩技术可以节省80%以上的费用,且一些难点问题通过压缩技术能够实现。 数据压缩技术种类繁多、应用广泛,技术不断发展,一些分支已成为当今研究的焦点。按照编码的失真程度数据压缩可以分为有损压缩和无损压缩。有损压缩即原始数据不能完全恢复,主要应用于图像和数字化的语音方面。无损压缩就是经过一个压缩后,能够产生和输入完全一致的压缩技术,主要用于存储数据库记录或处理文本文件。 2 目前主要文本压缩算法 文本压缩是根据一定方法对大量数据进行编码处理以达到信息压缩存储过程,被压缩的数据应该能够通过解码恢复到压缩以前的原状态,而不会发生信息丢失现象。发展到现在已经有很多关于文本压缩的算法,主要有Huff man 编码、算术编码等无损压缩和预测编码、量化、变换编码等有损压缩。如图1所示。 本文对常见的几种无损压缩算法进行综合分类,比较研究。 3 算法描述 3.1哈夫曼编码 美国数学家David Huff man 在上世纪五十年 代初就提出了这种编码,其主导思想是根据字符出 现的概率来构造平均长度最短的编码,并且保持编码的唯一可解性。也就是说,在源数据中出现概率越高的字符,相应码字越短;出现概率越小的字符,其码长越长,从而达到用尽可能少的码符号来表示源数据,达到压缩的效果。哈夫曼编码是一种变长的编码(因为其长度是随符号出现的概率而不同),在编码过程中,若各码字长度严格按照码字所对应符号出现概率的大小的逆序排列,则编码的平均长度是最小的。它最根本的原则是累计的(字符的统计数字3字符的编码长度)为最小,也就是权值(字符的统计数字3字符的编码长度)的和最小。 图1 文本压缩的分类 哈夫曼编码的编码过程如下:①对源信号的出现频率进行统计,每个源信号根据它出现频率大小被赋予一定的编码,高频率的信号对应短码,低频率的信号对应长码。 ②用信号对应的编码去取代源数据中的信号。在已知源数据流中各字符发生的概率情况下使用,即在压缩数据时,遵循固定的源数据符号代码表。在不允许两次扫描源文件的情况下,往往根据先验统计概率或假设构造这张代码表。压缩处理时,若源数据流中的符号与代码表中的符号相匹配,则用与之相对应的较短的代码替代该字符,否则不

实验报告-数据滤波和数据压缩实验

实验题目:使用Haar 小波和傅里叶变换方法滤波及数据压缩 1 实验目的 (1)掌握离散数据的Haar 小波变换和傅里叶变换的定义,基本原理和方法 (2)使用C++实现数据的Haar 小波变换和离散傅里叶变换 (3)掌握数据滤波的基本原理和方法 (4)掌握使用Haar 小波变换和离散傅里叶变换应用于数据压缩的基本原理和方法,并且对两种数据压缩进行评价 2 实验步骤 2.1 算法原理 2.1.1 Haar 小波变换 (1)平均,细节及压缩原理 设{x1,x2}是一组两个元素组成的信号,定义平均与细节为(12)/2a x x =+, (12)/2d x x =-。则可以将{a ,d}作为原信号的一种表示,且信号可由{a ,d}恢复,1x a d =+,2x a d =-。 由上述可以看出,当x1,x2非常接近时,d 会很小。此时,{x1,x2}可以近似的用{a}来表示,由此实现了信号的压缩。重构的信号为{a ,a},误差信号为 {|1|,|2|}{||,||}x a x a d d --=。因此,平均值a 可以看做是原信号的整体信息,而d 可 以看成是原信号的细节信息。用{a}近似的表示原信号,可以实现对原信号的压缩,而且丢失的细节对于最终信号的重构不会有重大影响。对于多元素的信号,可以看成是对于二元信号的一种推广。 (2)尺度函数和小波方程 在小波分析中,引入记号 [1,0)()() t X t φ=,其中, [1,0)() X t 表示区间[1,0]上的特征函 数。定义 ,()(2),0,1,...,21 j j j k t t k k φφ=-=- 称()t φ为Haar 尺度函数。由上式可知, ,()j k t φ都可以由 0,0() t φ伸缩和平移得到。 小波分析中,对于信号有不同分辨率的表示,当用较低分辨率来表示原始信号时,会丢失细节信息,需要找到一个函数来描述这种信息,该函数称之为小波函数。基本的小波函数定义如下:

相关主题