Chipalkatti et al.

– Estimating the Marketability Discounts 1

Estimating the Marketability

Discounts: A Comparison Between
Bid-Ask Spreads, and Longstaff’s
Upper Bound
Niranjan Chipalkatti, Carl Luft, Lawrence Levine, and Lummezen Mondal

predictive ability using restricted stock discounts. We offer

This paper contends that the discount for lack of marketability an alternate strategy for practitioners to estimate the DLOM
(DLOM) is the difference between the stock price of a liquid
which is based on existing market microstructure theory and
company and an equivalent illiquid company and reflects
the lack of a free-trading option that is embedded within a
literature.
company’s stock. Longstaff derived a model that views this This paper builds on the notion that the DLOM is essentially
liquidity swap as a lookback option. We equate this option the difference between the price of a liquid stock of a
to the Bid-Ask spread of a stock consistent with the market company and an equivalent illiquid stock.1 This difference
microstructure literature. We construct a model for the DLOM between the price of the illiquid stock and the price of an
using the Longstaff (1995) metric and the Bid-Ask spread of equivalent liquid stock arises because of the lack of a free-
Over-the-Counter Bulletin Board stocks as a proxy. We find trading option that implicitly comes along with the latter,
that our spread-based model does a better job of predicting Stoll (2000, p. 1482). In the market microstructure literature,
restricted stock discounts than the Longstaff metric. We include
the value of this free-trading option, or a liquidity option
a case study on two companies to illustrate our methodology.
associated with a publicly traded stock, has been equated to
the value of the Bid-Ask spread of a stock (Copeland and
nThe discount for lack of marketability (DLOM) is the Galai, 1983; Stoll, 2000; Bollen, Smith, and Whaley, 2004;
difference between the stock price of a liquid company and Chacko, Jurek, and Stafford, 2006). The Bid-Ask spread,
an equivalent illiquid company. In practice, the calculation i.e. the cost of making a market in a stock, represents an
of the DLOM is a subjective process involving many option written by the market maker that allows a liquidity
assumptions and often degenerates to an ad hoc reduction trader to buy stock at the ask price and to sell stock at the
of a certain percentage from the estimated ‘marketable’ bid price and provides compensation to the market maker
value of the asset without any economic reasoning. The for providing liquidity and assuming the fixed costs of order
lack of any comprehensive theory on valuation discounts processing, inventory holding costs, and the costs of adverse
as well as the lack of any legal benchmark has contributed selection. As also suggested by Kasper (1997), Chipalkatti
considerably to this state of affairs. The objective of this (2001), and Damodaran (2002, 2005), this paper proposes
paper is to construct a model for the DLOM using the Bid- that the Bid-Ask spread, estimated using data for equivalent
Ask spread of publicly traded liquid stocks and to validate its public company stock, is a viable market-based proxy for the
DLOM of an illiquid company. Consistent with the notion
Niranjan Chipalkatti is a Professor of Accounting at Seattle University, that the spread represents the premium on a free-trading
Albers School of Business & Economics in Seattle, WA. Carl Luft is option, we observe that it increases with volatility and
an Associate Professor of Finance in the Kellstadt Graduate School of
Business of DePaul University in Chicago, IL. Lawrence M. Levine is a trading interval similar to results previously documented in
Partner, Financial Advisory Services at McGladrey and Pullen in Chicago,
IL. Lummezen Mondal is a Manager, Financial Advisory Services at 1
Equivalent in this case implies identical cash flow streams and cost of
McGladrey and Pullen in Chicago, IL. capital.           

1
2 Journal of Applied Finance – No. 2, 2012
the market microstructure literature. An alternate model that discounts, the studies’ small sample sizes, and the lack of any
has received increasing attention as a viable alternative to theory relating to the DLOM have all resulted in considerable
estimate the DLOM (Damodaran, 2005; Duffy, 2009; Dyl friction between the internal revenue service (IRS) and
and Jiang, 2008; Stockdale, 2008) is the metric derived by taxpayers about the appropriate valuation discounts. In
Longstaff (1995). The Longstaff metric models the DLOM recent years, the Tax Court has considerably pared down the
using a lookback (call) option that views the DLOM as a DLOM to as low as 20 % in the McCord v. Commissioner
liquidity swap. case and has rejected the pre-IPO studies to be biased and
The objective of this paper is to construct a predictive unreliable.2 Primarily relying on the Bajaj et al. (2001)
model for the DLOM using the Bid-Ask spread of Over- study, the Tax Court in the Gross vs. Commissioner Estate
the-Counter Bulletin Board of Heck vs. Commissioner, MCord
(OTCBB) stocks and to calibrate This paper provides an alternative and Lappocases has significantly
the performance of this model to the Longstaff option method of pushed back against marketability
with that of the Longstaff metric
using restricted stock discount
estimating the discount for lack discounts that are deemed too high
(Robak, 2004; McMarrow, 2004)
data. To build a predictive of marketability, DLOM, for or inadequately supported.3, 4, 5
model for the Bid-Ask spread restricted stocks. In this post-Mandelbaum
we use data extracted from the environment, Pratt (2000) indicates
OTCBB market which has a high concentration of small that there will be a growing expectation that business
market capitalization companies that are relatively illiquid appraisers develop improved methodologies to quantify
(Levine and Luft, 2004). We find that the spread-based marketability discounts.6 This paper attempts to provide
model outperforms the Longstaff metric when it comes to such an improved methodology that is based on market
predicting the restricted stock discounts. microstructure theory and uses observable market data to
The paper develops as follows. The next section provides model the DLOM.
an institutional context to our research. We also review
the prior literature relating to the DLOM and the Longstaff B. Longstaff Metric
metric in this section. Section II presents the intuition
behind the notion that spreads and, therefore, the DLOM Longstaff (1995) theoretically modeled the DLOM and
can be modeled as free-trading options on the underlying derived the upper bound as a liquidity swap for a hypothetical
stock. In Section III, we construct a predictive model for investor who owned a lookback option and was assumed to
the DLOM using the Bid-Ask spread of OTCBB stocks. In possess perfect market timing. The Longstaff theoretical
Section IV, we test the predictive ability of our spread-based model holds that the DLOM is positively related to the
model and the Longstaff metric using actual restricted stock current value of the security, the length of the marketability
discounts to evaluate the performance of the metrics. Our restriction imposed on the restricted stock, and variance
concluding observations are provided in Section V. of stock returns. Finnerty (2002) expanded on Longstaff’s
model and demonstrated an additional negative relationship
I. Background with the dividend yield of the stock.
According to Longstaff (1995, p. 1772), the upper bound
A. Institutional Context derived from his model of DLOM also can be viewed as the
maximum amount an investor is willing to pay for immediacy
Most valuation practitioners and the Tax Courts continue in order to liquidate a security position and, hence, provides
to rely on the traditional empirical studies that attempt to an endogenous measure of the largest possible transaction
estimate the DLOM using data from the public markets. cost or the Bid-Ask spread for a security. Longstaff compares
These empirical studies include, (a) pre-initial public this transaction cost to the DLOM for restricted stocks and
offering (pre-IPO) discount studies, (b) studies that compare
market multiples of public versus private companies, and (c) 2
McCord v. Commissioner, 120 T.C. No. 13 (2003).
restricted stock studies. The pre-IPO studies, popularized           
by the work of John Emory and summarized by Pratt (2000, 3 Gross v. Commissioner, No. 99-2239/2257, 6th Cir. (2001).
p. 408), suggest that the discounts range from 25% to 45%.           
Using private placement data, however, the work of Bajaj, 4 Estate of Heck, T.C. Memo 2002-34 (2002).
Denis, Ferris, and Sarin (2001) concluded that the DLOM           
5
Lappo v. Commissioner, T.C. Memo 2003-258 (2003).
should be around 7.23%. Block (2007) observes an average           
discount of 20% to 25% with a low of 9% for financial firms. 6 Bernard Mandelbaum, et al. v. Commissioner, T.C. Memo 1995-255 (June
The age of some of these studies, the range and size of these 12, 1995).           
Chipalkatti et al. – Estimating the Marketability Discounts 3

observes that his metric closely approximates the observed the stock and simultaneously holds a put option to sell the
discount for lack of marketability. stock. In this case, the holder of the illiquid stock is essentially
buying a strangle option position for the underlying stock.9
II. The Bid-Ask Spread and the DLOM This assumes that there is a rational, unbiased agent willing
to make a market in the stock and that there is a buying and
Stoll (2000) proposes that the spread is a measure of the a selling interest in the stock. The cost of eliminating the
“friction” in financial markets that measures the difficulty illiquidity associated with the illiquid stock is equivalent
with which an asset is traded; the greater the friction, the to the premium on the strangle option position for such a
greater the spread.7 He suggests that the spread exists to stock which is also equivalent to the hypothetical Bid-Ask
compensate suppliers of immediacy for the free-trading spread (BAS) that would be charged on such a stock. In
or liquidity option they grant to the rest of the market. other words:
Previously, Copeland and Galai (1983) had theorized that
the market maker provides liquidity by giving a prospective PLIQ = PILLQ + BAS, (2)
trader a call option to buy stock at the asking price and a
put option to sell stock at the bid price. Thus, the cost of Therefore, equating (1) to (2),
obtaining immediacy in a publicly-traded stock trade is the
DLOM = BAS. (3)
size of the Bid-Ask spread, or, the premium on this strangle
option position written by the market maker, where the Consistent with this notion of the BAS as a strangle option,
option premium recovers the underlying cost of providing there is extensive documented evidence in the literature of a
these market making services to the investing public.8 positive association between the spread and the volatility of
Bollen et al. (2004) extend Copeland and Galai’s (1983) the returns and also the trading interval. We use the Bid-
work by assuming that the spread represents the premium Ask spread of OTCBB stocks to estimate the value of the
on a slightly in-the-money strangle option. More recently DLOM and the premium on the free-trading option of an
Chacko, Jurek, and Stafford (2006) also use an option-based illiquid stock because these OTCBB firms are relatively
framework to model liquidity. They suggest that when small in size, are illiquid, and have large trading intervals.
immediate exercise is desired the optimal option strike price In the next section, we outline the methodology used to
is the effective Bid-Ask spread. estimate spreads for OTCBB stocks to estimate the Longstaff
We propose that the Bid-Ask spread of publicly traded DLOM metric.
stock can be used to estimate the value of the DLOM and
the theoretical value of the free-trading option in the case III. Empirical Analysis
of illiquid stock. The DLOM represents the difference that
arises between the price of a liquid stock and an equivalent A. Metrics
illiquid stock because of the lack of a free-trading option
We estimate the two metrics discussed above: Bid-Ask
that implicitly comes along with the ownership of a stock
Spread and the Longstaff Option, as percentages. The Bid-
that has a liquid market. Assume that PLIQ is the price
Ask spread is calculated as:
of liquid stock and PILLQ is the price of illiquid stock of
an equivalent company. The relationship between PLIQ ,  
PILLQ, and DLOM, the value of the discount for lack of  (Ask − Bid) 
BidAsk% =  *100.
marketability is: (Ask + Bid) 
 
PLIQ = PILLQ + DLOM. (1)  2  (4)

The (relative) illiquidity that comes with such stock may be The spread is the difference between the price a market
“eliminated” if the owner holds a call option to buy more of maker is willing to pay, the BID, and the price the market
maker is willing to accept, the ASK, as a percentage of the
spread’s midpoint.
7
The microstructure literature documents that the Bid–Ask spread is set
by the market maker to recover three sets of costs: order processing costs,
inventory holding costs due to the need to hold sub-optimal levels of The Longstaff metric is estimated as:
inventory and asymmetric information or adverse selection costs of trading
with informed investors.
           9
The option’s maturity is equal to the holding period of stock. The value of
8
A strangle position is created when a put option and call option are this option position increases with the volatility of the underlying private
combined, but the exercise prices differ for the call and put options. In stock as well as the time it takes to liquidate a position in that stock. The
this case, the call option exercise price equals the dealer’s asking price, value of the option position is also a function of the price of the underlying
and the dealer’s bid price defines the exercise price for the put option. stock.           
          
4 Journal of Applied Finance – No. 2, 2012

terms for the computation of the Longstaff metric. Next, we

 σ 2T   σ 2 T  σ 2T  σ 2T 
F  V2   N  V exp     V. (5) calculated each company’s monthly share price volatility
 2   2  2π  8  via the Garman-Klass-Parkinson (GPK) methodology as
modified by Bilson (2003).11 This methodology is well suited
Where: to our data since it allows us to capture the monthly variation
F = the value of the discount for lack of marketability, in each firm’s share price even though many of these firms
V = the current price of the asset, did not trade on a daily basis. The Garman-Klass-Parkinson
σ2 = the monthly volatility of the asset price changes, volatility was calculated via:
T = the trading interval expressed in monthly terms.
N( ) = the cumulative normal distribution function. GKPσ2 = [ln(Ot/Ct-1)]2 + .5[ln(Ht/Lt)]2
- .31[ln(Ct/Ot)]2 (6)
The Longstaff metric is expressed as a percentage by
dividing the option value, F, by the underlying asset value, V. Where:
GKPσ2 = the monthly volatility for the firm’s share prices,
B. Data O = the firm’s opening share price at the beginning of the
To estimate the metrics, daily data for 1,850 individual month,
OTCBB firms traded between December of 1995 and C = the firm’s closing share price at the end of the month,
December of 2000 were obtained from Nasdaq. We did not H = the firm’s highest share price during the month,
include OTCBB data reported after December 31st 2000 L = the firm’s lowest share price during the month,
because the nature of the market changed in 2001. After t = time.
January 1st, 2001, OTCBB listed firms were required to file After estimating each firm’s monthly share price volatility,
their financial statements with the Securities and Exchange we computed each firm’s market capitalization and trading
Commission (SEC). We felt the OTCBB firms listed through activity for each day between December of 1995 and
December 31st of 2000 comprised a better benchmark for December of 2000. The market capitalization value was
illiquid firms The inclusion of firms that chose to stay on the determined by multiplying the firm’s closing price on that
OTCBB after this regulation will create a self-selection bias day by the number of shares outstanding.12 Monthly trading
in the sample as improved disclosure requirements enhanced activity for each firm was computed by summing its daily
the underlying liquidity of their stock. The data needed to trading activity during the month and then dividing the
estimate the metrics included the closing Bid price, the total number of shares traded by the total number of shares
closing Ask price, the transaction price, and the trading outstanding at the end of the month.
volume.10 The monthly number of shares outstanding and
any splits or dividends that occurred were obtained from C. Portfolio Formation
Bloomberg.
Once each firm’s trading history had been established It is well documented in the market microstructure
from December of 1995 through December of 2000, we literature (see Stoll (2000) for a comprehensive review),
transformed the data and computed the percentage spread that the Bid-Ask spread is significantly related to trading
and the Longstaff metric. We estimated the trading interval activity, volatility, market capitalization and trading interval.
for every trade by determining the number of calendar days We assign the OTCBB firms into 24 portfolios based on
that elapsed between trades. We then transformed each the firm-specific value for each of these four dimensions
observation’s trading interval into both monthly and annual relative to the overall sample average. The classification
procedure was a four step iterative process which employed
the monthly median values for market capitalization, trading
10
When we examined the data it was obvious that we had three problems.
First, there were days when Bid and Ask values were not reported. Second, activity, share price volatility, and the monthly mean value
there were days when the trading volume was zero. Third, there were for trading interval. Firms were identified as High Market
days where we observed trading volume, a Bid price, an Ask price, but Cap firms if their market capitalization value exceeded the
no transaction price. We addressed the first problem by assigning the most
recent Bid and Ask values to the observations that were missing Bid and Ask
values. If there were several successive days where the Bid and Ask values 11
This volatility measure is based on the firm’s open, high, low, and closing
were not reported, then we continued to use the most recently observed prices over a given trading interval and uses as much of the share price
values for the Bid and Ask values. We dealt with the second problem by information as possible.
defining a good trading day as one where trading volume was reported. If           
no trading volume was reported, then we dropped the observation. The third 12
If a trade did not occur on that day, then we used the average of the most
problem was resolved by using the mid point of the Bid-Ask spread as the recent trading day’s Bid-Ask spread as the price in the market capitalization
transaction price. Thus, we established the trading history for each firm. calculation.           
          
Chipalkatti et al. – Estimating the Marketability Discounts 5

Table I. OTCBB Firm Portfolio Classification

Over the Counter Bulletin Board Firm portfolios created via four dimensions: market capitalization, share price volatility, trading activity,
and trading interval. Each cell contains the average value the portfolio’s dimensions: market capitalization ($100,000); monthly trading
activity (Volume / Shares Outstanding) trading interval (days) GKP volatility (monthly sigmas); and the median values for the Bid Ask
spread and Longstaff percentages.
High monthly trading activity Low monthly trading activity
Long Medium Short Long Medium Short
trading trading trading trading trading trading
interval interval interval interval> 5 interval interval
> 5 Days 2 -5 Days < 2 Days Days 2 -5 Days < 2 Days

High Market High σ

Capitalization Portfolio # 1 2 3 7 8 9
Size $27,904 $14,931 $32,905 $37,283 $20,043 $32,415
Activity 1.39% 0.59% 0.95% 0.03% 0.04% 0.06%
Interval 14.1 2.1 1.5 13.4 2.4 1.6
Volatility 95% 43% 37% 59% 38% 34%
BidAsk% 9% 12% 6% 28% 17% 9%
Longstaff % 17% 8% 7% 14% 8% 6%
Low σ
Portfolio # 4 5 6 10 11 12
Size $24,778 $25,182 $37,901 $116,658 $78,408 $89,460
Activity 0.56% 0.48% 0.60% 0.04% 0.04% 1.20%
Interval 9.0 2.4 1.5 8.4 2.5 1.6
Volatility 7% 11% 15% 7% 10% 13%
BidAsk% 7% 7% 4% 8% 7% 6%
Longstaff % 2% 2% 3% 2% 2% 3%
Low Market High σ
Capitalization Portfolio # 13 14 15 19 20 21
Size $1,713 $2,549 $3,228 $2,104 $2,945 $3,559
Activity 1.20% 1.52% 5.66% 0.04% 0.05% 0.06%
Interval 10.0 2.3 1.5 10.3 2.6 1.6
Volatility 59% 47% 43% 57% 44% 41%
BidAsk% 39% 22% 12% 47% 26% 17%
Longstaff % 16% 10% 8% 17% 9% 8%
Low σ
Portfolio # 16 17 18 22 23 24
Size $2,751 $3,231 $3,869 $2,927 $3,504 $4,144
Activity 0.88% 0.84% 1.99% 0.05% 0.05% 0.06%
Interval 8.2 2.4 1.5 8.0 2.6 1.6
Volatility 11% 14% 16% 12% 14% 16%
BidAsk% 18% 12% 9% 21% 14% 12%
Longstaff % 4% 3% 3% 4% 3% 3%

median and Low Market Cap firms otherwise. The same We examine the behavior of the Bid Ask spread and the
logic was used to classify firms according to trading activity Longstaff metric across the four dimensions to confirm that
and volatility. the portfolio median values for the spread and the Longstaff
In the case of trading interval, we employed three metric behave in a manner that is consistent with the results
categories instead of two. Firms with mean trading intervals previously obtained for effects related to firm size, trading
that were less than or equal to two days were placed in the activity, trading interval, and volatility. Table II displays the
Short Trading Interval group and those with mean trading results.13 We observe that the sample behaves in a manner
intervals between two days and five days were classified as
having Medium Trading Intervals. All others were classified 13
We tested the two metrics via the Wilcoxon nonparametric test and
as firms that have Long Trading Intervals. The final results rejected the null hypothesis of equal median values at the .0001 level for all
with portfolio averages for each metric are portrayed in the median values associated with the metrics. Thus, the two metrics appear
Table I. to generate different values across all four dimensions.           
 6 Journal of Applied Finance – No. 2, 2012

Table II. Size, Trading Activity, Volatility and Trading Interval Effects across OTCBB Portfolio Medians
for Bid-Ask Spread Percentage Metric and Longstaff Metric
Over the Counter Bulletin Board Firm portfolios created via four dimensions: market capitalization, share price volatility, trading activity,
and trading interval. Each cell in the second column contains the average value the portfolios’ percentage bid-ask spread, Equation (4),
and the cells in the third column present the average values, across the portfolios, for the Longstaff metric, Equation (5).
Bid-Ask Longstaff
Compare size effect
Large cap portfolios 1:12 6.90% 4.42%
Small cap portfolios 13:24 17.47% 6.90%
Compare activity effect
High activity portfolios 1:6 & 13:18 8.28% 5.73%
Low activity portfolios 7:12 & 19:24 11.92% 4.92%
Compare volatility effect
High sigma portfolios1:3 7:9 13:15 19:21 13.27% 7.92%
Low sigma portfolios 4:6 10:12 16:18 22:24 7.48% 2.85%
Compare trading interval effect
Long interval portfolios 1,4,13,16,7,10,19,22 21.39% 6.11%
Medium interval portfolios 2,5,14,17,8,11,20,23 14.60% 5.38%
Short interval portfolios 3,6,15,18,9,12,21,24 7.52% 5.15%

that is consistent with published results in the market measured on a daily basis as the current share price times the
microstructure literature. number of outstanding shares for each portfolio p, in year y,
Large firms exhibit smaller spreads than small firms; and for month m.
firms that experience higher
0  trading activity have smaller
p,y,m   .average of the log of the
2
ln(BidAsk%) p,y,m 1 ln(MC) p,y,m  2 ln(L) p,y,m  3 ln(GKPσ )p,y,m =the 4 ln(I)
monthly
spreads. The spreads for high volatility firms are larger than
Garman-Klass-Parkinson monthly volatility for each
the spreads for low volatility firms and the longer the trading
portfolio p, year y, for month m.
interval, the greater the spread. The results for Longstaff’s
Ask%)p,y,mmetric 1 ln(MC)
 0 also ln(L)the
p,y,m  2 with
are consistent  3 ln(GKPσ
p,y,m market
2
)p,y,m  4 ln(I)p,y,m =the
microstructure . monthly average of the log of the trading
literature for the size, trading interval, and the volatility interval expressed as days since the last trade for each
effects. However, the trading activity effect is contrary to portfolio p, in year y, for month m.
our expectation that marketability costs will decrease with
greater trading volume. The two
ln(BidAsk%) metrics differ only with
p,y,m  0  1 ln(MC) p,y,m  2 ln(L) p,y,m =the 3 ln(GKPσ
2
monthly)p,y,m  4 ln(I)
average of the   .of the daily
p,y,mlog
respect to their association with the trading activity variable. trading activity calculated as daily number of shares traded
We model the portfolio averages for the Bid-Ask spread as as a percentage of outstanding shares for each portfolio p, in
a function of firm size, trading activity, price volatility, and year y, for month m.
trading interval. The model is specified as follows: The results are presented in Table III.
Once again, we observe that our sample behaves in a
ln(BidAsk%)p,y,m  0  1 ln(MC)p,y,m  2 ln(L)p,y,m  3 ln(GKPσ 2
fashion)that 
p,y,m is   . the relationships documented
4 ln(I)p,y,mwith
consistent
previously in the market microstructure literature. There
ln(L)p,y,m  3 ln(GKPσ2 )p,y,m  4 ln(I)p,y,m   . is a significant inverse relationship between the percentage
(7)
Bid-Ask spread and the size and trading activity variables.
Where: Also there are significant positive relationships between the
percentage Bid-Ask spread and volatility and also trading
4 ln(I)p,y,m  power
2
ln(BidAsk%)
 the
p,y,m = 1 ln(MC)
0 monthly 2 ln(L)
p,y,m value
average of the 3 ln(GKPσ
 of
p,y,mlog interval.)p,y,m
The explanatory . of this regression, as
the percentage Bid-Ask spread, for each portfolio p (1 to 24), measured by the adjusted R , is 89%.14 2

in year y (1995 to 2000), for month m (1 to 12).

p,y,m   .In his paper, Longstaff suggests that his metric can also be considered to
2
 0  1 ln(MC)p,y,m =the
2 ln(L) 3 ln(GKPσ
p,y,m average
monthly of the )log 4 ln(I)
 the
p,y,mof market 14

capitalization be a measure of the largest Bid-Ask spread for a security. We check

 Chipalkatti et al. – Estimating the Marketability Discounts 7

Table III. Regression Model: ln(%Bid Ask Spread)=f(ln(Market Cap), ln(Liquidity),ln(GKP volatility),
ln(Trading Interval)
The following regression model was estimated:

ln(BidAsk%)p,y,m  0  1 ln(MC)p,y,m  2 ln(L)p,y,m  3 ln(GKPσ2 )p,y,m  4 ln(I)p,y,m   .

ln(BidAsk%)p,y,m =  0 
the 1 ln(MC)
monthly average  2 ln(L)
p,y,m value log 
of thep,y,m 3 ln(GKPσ
ofthe
2
percentage)p,y,m ln(I)p,y,m
 4spread,
Bid Ask  . portfolio p (1 to 24), in year y
foreach
(1995 to 2000), for month m (1 to 12),

p,y,m   . measured on a daily basis as the current share price times the
2
)p,y,m  0  1 ln(MC)p,y,m = the
2 ln(L)
monthly  3 ln(GKPσ
p,y,maverage )p,y,m
of the log of the  4 ln(I)
market capitalization
number of outstanding shares for each portfolio p, in year y, for month m,

ln(L)p,y,m  3 ln(GKPσ2 )p,y,m =the

4 ln(I) p,y,m 
monthly .
average of the log of the Garman-Klass-Parkinson monthly volatility for each portfolio p, year y, for
month m ,

KPσ2 )p,y,m  σln(I)p,y,m   = the monthly average of the log of the trading interval expressed as days since the last trade for each portfolio p, in year
y, for month m,

)p,y,m  0  1 ln(MC)p,y,m =  2 ln(L)

the monthly  3 ln(GKPσ
p,y,maverage
2
of the log )of 4 ln(I)
thedaily
p,y,m trading   . calculated as daily number of shares traded as a percentage of
p,y,mactivity
outstanding shares for each portfolio p, in year y, for month m.

Equation (7)
Dependent variable: Monthly average ln (Bid Ask %)p,y,m
Over the Counter Bulletin Board firms classified by portfolio, year, and month.
Variable Coefficient t-statistic
Constant 1.8338 20.87 ***
Average ln(MC) -.2832 -50.00***
Average ln(L) -.1129 -25.94***
Average ln(GKPσ2) .4067 47.13***
Average ln(I) .5891 47.63***
Observations 1440
Adjusted R2 89%
***Significant at the 0.01 level

IV. Prediction of Restricted Stock Ask spreads (Table II) provide some insights on DLOM
Discounts using the Bid-Ask Spread and values. The median spread for firms with long trading
intervals is about 21% as compared to 7.5 % for those
the Longstaff Metric with short trading intervals. The spread for low market
capitalization portfolios with low trading activity, portfolios
In the previous section, we proposed that we can use the
19 to 24,is approximately 19%. The spread for this category
spread of OTCBB stocks as proxies for estimating the
of portfolios with long trading intervals range from 22% to
DLOM of illiquid stock. The data for the OTCBB Bid-
35% and the spread for portfolios with short trading intervals
range from 9% to 15%. In this section we test the predictive
whether using his metric would yield a better predictive model of the performance of the Bid-Ask spread metric as a measure for
spread. We regress the observed Bid-Ask spread as a function of firm size,
the DLOM. We mimic Longstaff (1995) and Finnerty (2002)
trading activity, and the Longstaff metric. We use the monthly average
value of the log of the Longstaff metric for each portfolio p, in year y, and calibrate the performance of the Longstaff metric and a
for month m. The results indicate that the percentage Bid Ask spread is Bid-Ask spread based metric on a sample of restricted stocks
significantly and inversely related to firm size and trading activity. The discounts.
Longstaff metric is also significant and positive. The adjusted R2, is 81%
The sample of restricted stock transactions was obtained
and less than that obtained using the trading interval and the volatility
variables as a linear function. The significant coefficients for Market from FMV Opinions Inc. who conduct annual surveys of
Capitalization and Trading Activity indicate that the the Longstaff metric restricted stock transactions. The FMV price data include
          
8 Journal of Applied Finance – No. 2, 2012

the transaction price for the restricted stock’s sale; the the parameters to predict the DLOM by re-estimating
number of shares outstanding; the monthly trading volume Equation (7) after making one adjustment. We dropped all
for the exchange traded shares; the monthly Open, Close, OTCBB observations with trading intervals less than sixty
High, and Low prices for the exchange traded shares; the days as this eliminated much of the noise associated with
prior month’s average of the observed high and low values the short trading interval data.19 We felt that the OTCBB
for the restricted stock’s exchange traded shares, the stock’s observations with trading intervals that exceeded sixty days
annualized volatility measured as the standard deviation more closely matched the long trading interval of either
of the prior year’s weekly returns; and the stock’s required one year or two years of the restricted stock data. To be
holding period, expressed in years, based on the SEC ruling consistent, we dropped the trading interval variable from our
at the time of the restricted stock transaction. We used these estimating equation. This left us with 48,108 daily trading
data to calculate each restricted stock’s Garman-Klass- observations that we had to place in portfolios so that we
Parkinson volatility measure and each restricted stock’s could calculate monthly averages for the log values of the
percentage discount.15 percentage Bid-Ask spread, market capitalization, trading
We tested the Longstaff metric and our Bid-Ask spread activity, and volatility. We followed the same portfolio
model of the DLOM on a set of restricted stock discount classification procedure as was described previously. Next,
transactions that occurred between March of 1991 and we obtained monthly average values for each variable of
November of 2000. The Longstaff metric and our Bid- interest for each portfolio. This procedure yielded 336
Ask spread model were tested twice: once using the FMV monthly portfolio observations, containing 511 firms,
volatility, and once using the Garman-Klass-Parkinson which were used in a regression to obtain the parameters we
volatility measure. The first test, which employed the needed to estimate the DLOM for the restricted stock.20 The
FMV volatility measure, was performed on a sample of regression equation was specified as:
338 restricted stock discount transactions; the second test 2
used the Garman-Klass-Parkinson volatility measure and ln(BidAsk%)p,m  α0 + α1 ln(MC)p,m  α2 ln(L)p,m  α3 ln(GKPσ p
was performed
ln(BidAsk%) on 142
p,m of
 αthe 338 restricted stock discount
0 + α1 ln(MC)p,m  α 2 ln(L)p,m  α3 ln(GKPσ p,m   .
2
(8)
transactions.16
We used both the percentage Bid Ask spread and the Where:
Longstaff metric (Equation 5) to predict the marketability
2
discount for the restricted stockwhich we then compared ln(BidAsk%)p,m =αthe 0 + monthly
α1 ln(MC) p,m  α 2 ln(L)p,m  α3 ln(GKPσ p,m
average value of the log
to the actual restricted stock discounts.17 To compute the of the percentage Bid-Askspread, for each portfolio p, in
Longstaff metric for each restricted stock observation, we month m where there are 8 portfolios, and 42 months.
set V, the underlying asset value, to be equal to the current
transaction price of the stock and ln(BidAsk%) p,m  α0 +
used monthly volatility. 18 α ln(MC)
1 =α
p,m  2 ln(L)
the α3 ln(GKPσ
p,m average
monthly
2
 . market
p,mofthe
of the log
We set T, the trading interval, to be equal to the number of capitalization measured on a daily basis as the current
months contained in the required holding period. share price times the number of outstanding shares for each
In the case of the Bid-Ask spread based model, we obtained portfolio p, in month m,

15
Weln(BidAsk%) p,m  α0as+the
calculated the discounts α1percentage
ln(MC)p,m  α2 ln(L)
difference between  α3 ln(GKPσ p,m 
p,m the
2
= the
. monthly average of the log of the
prior month’s average of the exchange traded stock’s observed high and Garman-Klass-Parkinson monthly volatility for each
low values and the restricted stock’s transaction price. portfolio p, in month m,
          
Our sample of 142ln(BidAsk%)
restricted stock observations
p,m  α0 + was defined by
α1 ln(MC) p,mthe
 α2 ln(L)p,m = αthe
3 ln(GKPσ
2
p,m   .of the log of the daily
16
monthly average
availability of the price data required to compute the Garman-Klass-
Parkinson volatility: monthly Open, Close, High, and Low values for trading activity calculated as daily number of shares traded
exchange traded shares issued by firms which also completed restricted as a percentage of outstanding shares for each portfolio p, in
stock transactions between 1991 and 2000. month m.
          
17
Unfortunately, we were unable to obtain financial statement data and
dividend data for most of the OTCBB stock. This would have permitted us 19
130 of the restricted stock observations had one year trade intervals, and
to include other factors that impact marketability like dividend, profitability,
the remaining 208 observations had two year trade intervals.
etc. in our spread regression equation. This is the approach taken by
          
Damodaran (2002) and Chipalkatti (2001). 20
There were 480 possible monthly portfolio observations where trading
      intervals exceeded five days, 8 portfolios over the five year, 60 month,
18
The FMV volatility was used to calculate the discount for all 338 discount period; but only 42 months between January of 1995 and December of 2000
observations. The Garman-Klass-Parkinson (GPK) volatility also was used contained observations for the eight portfolios with trading intervals greater
to obtain the discounts for the subset of 142 of observations where the GKP than 60 days. This yielded 336 monthly portfolio observations with trading
volatility could be obtained.            intervals greater than 60 days.           
 Chipalkatti et al. – Estimating the Marketability Discounts 9

Table IV. Regression Results for Monthly Average Natural log of the Percentage Bid Ask Spread for
Trading Intervals Greater Than 60 Days:
ln(%Bid Ask Spread)=f(ln(Market Cap), ln(GKP volatility), ln(Liquidity)
The following regression model was estimated:
ln(BidAsk%)p,m  α0 + α1 ln(MC)p,m  α2 ln(L)p,m  α3 ln(GKPσ2 p,m   p,m .
ln(BidAsk%)p,m = α 0+
the α1 ln(MC)
monthly average  α2 of
p,mvalue ln(L) p,m 
the log α3 ln(GKPσ
of the
2
p,m
percentage Bid  spread,
Ask . for each portfolio p, in month m, where
there are 8 portfolios and 42 months,

2 ln(L)p,m  α3 ln(GKPσ p,m   .

%)p,m  α0 + α1 ln(MC)p,m = αthe monthly average of the log 2of the market capitalization measured on a daily basis as the current share price times the
number of outstanding shares for each portfolio p, in month m,

ln(L)p,m  α3 ln(GKPσ2 p,m   . monthly average of the log of the Garman-Klass-Parkinson monthly volatility for each portfolio p, in month m,
= the

n(MC)p,m  α2 ln(L)p,m  α3 monthly

= the ln(GKPσ 2
p,mof
average  . log of the daily trading activity calculated as daily number of shares traded as a percentage of
 the
outstanding shares for each portfolio p, in month m.
Equation (8)
Dependent variable: Monthly average ln (Bid Ask %)p,m
Over the Counter Bulletin Board firms classified by portfolio and month.
Variable Coefficient t-statistic
Constant 3.6906 13.06***
Average ln(MC) -.3737 -20.42***
Average ln(L) -.1632 -8.36***
Average ln(GKPσ2) . 2921 11.43***
Observations 336
Adjusted R2 72%
***Significant at the 0.01 level

Table IV presents the results from estimating Equation 8 the restricted stock discounts calculated from the FMV data.
and shows that 72% of the variation in the percentage Bid- The results are presented in Table V.
Ask spread is explained by the regression. Furthermore, the Table V contains the descriptive statistics for the observed
significant relationships detected earlier in Table III still restricted stock discounts and for the estimated discounts
hold: The percentage Bid-Ask spread is inversely related to generated by the Longstaff model and the Bid-Ask spread
both firm size and trading activity and positively related to model. Table V also reports two measures of forecast
volatility. accuracy: The Mean Square Error and Theil’s U-Statistic
The monthly average of the percentage Bid-Ask spread for both models. Panel A contains the results for the 338
in equation 8 is our spread-based proxy for the DLOM. discount transactions that used only the FMV volatility.
The coefficients generated by Equation 8 were then used Panel B presents the results for the smaller sample of 142
to estimate the DLOM for the restricted stock discount observations with two estimated discount values- one using
observations. The spread-based DLOM was also estimated the FMV volatility measure and the other using the GKP
twice, once using the FMV volatility for 338 discounts, volatility measure.
and then using the GKP volatility for a sub-sample of 142 According to Panel A the median percentage discount for
discounts for which the latter could be calculated. the entire sample was approximately 19%. Furthermore,
Since we need to measure DLOM for sell-side restrictions both the Longstaff model and the Bid-Ask spread model
only, we transformed the calculated DLOMs into half generate higher percentage discounts than the observed FMV
spreads. This transformation makes the calculated DLOMs percentage discounts. The Bid-Ask spread model yields
consistent with Longstaff spreads.21 These were compared to median DLOM of approximately 28% as compared to 106%
obtained by the Longstaff metric. Finally, the Bid –Ask
spread model appears to be superior to the Longstaff model
21
The spread represents the cost of providing both buy-side and sell-side as an estimator of the percentage restricted stock’s DLOM.
liquidity. Longstaff’s metric, on the other hand, measures the DLOM when
there is a restriction to sell. The fact that the spread works out to be almost The results in Panel B are very similar. These results are
double the Longstaff metric (See Table II) makes sense in this context. not surprising when one considers the fundamental nature of
          
10 Journal of Applied Finance – No. 2, 2012

Table V. Descriptive Statistics for Restricted Stock DLOM Percentages Calculated Via: the FMV Data,
the Longstaff Option Model, and the Bid-Ask Regression Model
The statistics are for the percentage half spreads observed in the FMV data, and also estimated via the Longstaff option metric, and
Equation (10), where Equation (10) employs the parameters generated by the portfolio Bid Ask regression model, Equation (9). Panel A is
the set of 338 observed discounts that employed only the FMV volatility. Panel B is the set of 142 observed discounts that had sufficient
data to employ both the GKP and FMV volatilities.

Observed Longstaff Estimates Bid Ask Estimates

Restricted FMV Volatility GKP FMV GKP
Stock Volatility Volatility Volatility
Discounts
Panel A. Restricted Stock Discounts Based on FMV Volatility
Number of 338 338 338
Observations
Mean .2277 2.1719 .3592
Median .1871 1.0660 .2882
Sigma .1621 11.6340 .2219
Maximum .7949 205.9746 1.3951
Minimum .0005 .2377 .0538
Mean Square Error 138.0211 .0656
Theil’s U 42.057 .917
Panel B. Restricted Stock Discounts Based on Both FMV and GKP Volatilities
Number of 142 142 142 142 142
Observations
Mean .2188 3.1589 3.2421 .3277 .3317
Median .1733 1.0274 .9927 .2645 .2635
Sigma .1572 17.8448 13.2215 .2125 .2234
Maximum .7059 205.9745 114.7837 1.3597 1.2478
Minimum .0017 .2635 .1662 .0741 .0604
Mean Square Error 323.318 181.2028 .0599 .0637
Theil’s U 66.822 50.025 .910 .938

each metric. The percentage Bid-Ask (half) spread is market In Panel A, the U-statistic of 0.917 obtained for the Bid-Ask
determined and captures the effects of inventory holding spread model dominates the U-statistic of 42.015 obtained
costs, liquidity costs, and asymmetric information costs on a for Longstaff’s model, and leads to the conclusion that the
firm’s share price. The Longstaff metric is an upper bound Bid-Ask spread model provides better and more accurate
for the DLOM and assumes equally well informed parties estimates of the discounts than does the Longstaff metric.23
where the share owner is prohibited from selling until a
specific date. If U equals one, then the forecast is only as reliable as the no change
The predictive ability of the estimated DLOMs generated extrapolation. If U is greater than one, then the forecast is worse than the no
by the two metrics was assessed via Theil’s U-statistic. change extrapolation.           
A U-statistic which is less than one indicates that a model
performs better than a naive no change extrapolation model.22
23
When the FMV volatility and the GKP volatility results in Panel B are
compared for the Bid-Ask spread model, there does not appear to be much
difference in the estimated discounts. This is encouraging for the GKP
22
Theil’s U-statistic is a measure of forecast accuracy that is based on the volatility because this estimate is based on the prior month’s share price
mean square error and measures the severity of the forecast errors. If all movement: Open, Hi, Low, and Close; and does not require a year’s worth
forecasts are perfect, then U equals zero. If U takes a value which is greater of price and return data as does the FMV volatility. This observation was
than zero, but less than one, then the forecast is not perfect, but it is better confirmed by a t-test that failed to reject the null hypothesis of equal mean
than what can be achieved by using a naive no change extrapolation. values for the FMV and GKP discounts at the .1 level.           
Chipalkatti et al. – Estimating the Marketability Discounts 11

Panel B provides further evidence regarding the performance V. Case Study

of the two models. The Theil U-statistics generated for the
estimates made using the FMV volatility measure is 0.910 We illustrate the application of the Bid-Ask spread model
for the Bid-Ask spread model and 66.82 for the Longstaff using data for two companies. The first company, Carrington
model. The U-statistics based on the GKP volatility estimate Laboratories, was traded on the American Stock Exchange
is 0.938 for the Bid-Ask spread model and 50.025 for the under the ticker symbol CRN. The second firm, Genus
Longstaff metric. Both results reinforce the conclusion that Company, was larger than Carrington, and traded Over-
the Bid-Ask spread model outperforms the Longstaff metric. the-Counter under the ticker symbol GGNS. The raw data
Although these results show that the Bid-Ask spread model required to apply the model were extracted from the Center
generated better forecasts of the observed restricted stock for Research in Security Prices (CRSP) database and are
DLOM than the Longstaff model, we cannot reject the latter presented in Table VI for both companies. The transaction
based just on Theil’s U-statistic. Hence, we employ Davidson dates represent the first day of the month in which the
and McKinnon’s J-test (1981) for non-nested hypothesis to restricted stock was issued as listed in the FMV database.
test for the better, or the most acceptable, model specification We chose these companies to demonstrate the performance
for the prediction of restricted stock discount. The J-test is (in terms of DLOM) of the Bid-Ask spread model and the
implemented by specifying two hypotheses, H0 and H1 and Longstaff equation for two different firms based on market
then testing one against the other. For the observed restricted capitalizations, trading activity, and volatility. Comparing
stock discounts, the test specification is: the data in Table VI reveals that CRN was smaller than
GGNS, had a higher share price than GGNS, but had a much
H0: ln (2*Observed Discount) = α+β1ln(Size)+β2ln(Activity) lower trading volume than GGNS. The (GKPσ2 ) variances
+β3ln(Volatility)+ε, (9)24 also indicate that CRN’s share price was less volatile than
that of GGNS. Both the Bid-Ask spread model and the
H1: ln(2*Observed Discount) = β1ln(2*Longstaff Predicted Longstaff equation are applied to each firm on the appropriate
Value) +ε. (10) transaction dates. FMV reports the discounted values for
The J-test can have four possible outcomes. The results restricted stock issues on the first trading day of the month or
may suggest that both models are acceptable or that both the transaction date. Hence, we applied the two models using
are not acceptable. The results may also find one of the two the raw data as observed in the month prior to the transaction
models acceptable and the other model not acceptable. date. We assume that the discounts reflect activity during
We estimated both models and generated values for the this month. . The size, activity, and volatility data (see Table
estimated discount; FMVBAHAT from H0 and FMVLHAT VI) are used to generate the values for the model’s inputs:
from H1. The J-test involves testing the significance of the Market Capitalization (MC), Trading Activity (L), and
coefficient on FMVLHAT when included in the equation Garmon-Klass-Parkinson variance (GKPσ2). The values for
for H0 and then, testing the significance of the coefficient MC were obtained by multiplying the prior month’s average
on FMVBHAT when included in H1. If the coefficient on price by the number of shares outstanding during the prior
FMVLHAT (FMVBHAT) is not significant, the Bid-Ask month. The values for L were generated by dividing the
spread model (Longstaff model) is acceptable while the number of shares outstanding during the prior month by the
alternate model is not acceptable. The slope coefficient on prior month’s trading volume where the number of shares
FMVLHAT was not significant (t-statistic of -0.30 with a outstanding had to be expressed in hundred share units to
p-value of 0.764). This indicates that the Bid-Ask spread be consistent with the monthly trading activity. The GKPσ2
model was not rejected by H1, the Longstaff metric. On the values were calculated as illustrated using the appropriate
other hand, the coefficient on FMVBHAT was significant share price data.
(t-statistic of 17.93 at p < 0 .0001). The J-test results indicate Table VII provides the details of the calculations used to
that the Bid-Ask spread based model was acceptable while generate the restricted stock discounts (DLOM) for CRN
the Longstaff metric was not acceptable as a model for and GGNS employing the Bid-Ask Spread model and the
estimating restricted stock DLOM. Longstaff equation. The Bid-Ask spread model’s percentage
In sum, the J-test results and the results for Theil’s U discounts were generated by exponentiating the estimated
indicate that the Bid -Ask spread model is superior to the value of the dependent variable and then dividing by two
Longstaff option model for estimating the DLOM. to obtain the half-spread. The Longstaff discounts were
generated by inserting the monthly volatility values for
σ2, with T equal to 24 months, to yield the values for the
24
We multiplied the observed discount by two to convert half spreads in to
two terms in the equation. Then, the first two terms were
full spreads to be consistent.            added and one was subtracted from the result to derive the
12 Journal of Applied Finance – No. 2, 2012

Table VI. Data for Bid Ask Spread Model Application

Raw data extracted from Center for Research in Security Prices, CRSP, for Carrington Laboratories, CRN, and Genus Company, GGNS.
Used to illustrate the application of Bid-Ask spread model and Longstaff metric.

Company Ticker CRN GGNS

Exchange AMEX OTC
Transaction Date 10/1/1992 2/1/1995
Holding Period: Months 24 24
Size
Shares Outstanding in Thousands: Prior Month 6,744.000 12,813.000
Activity
Trading Volume in hundreds: Prior Month 1,943.00 35,392.00
Share Prices
Ot Opening Bid Price: First Day Transaction Month $11.875 $7.750
Ct Closing Bid Price: First Day Transaction Month $12.000 $8.125
Ct-1 Closing Bid Price: Prior Day $11.125 $7.750
Ht Monthly High: Prior Month $12.500 9.375
Lt Monthly Low: Prior Month $11.125 $7.500
Average Price: Prior Month $11.813 $8.438
Implied FMV discount 23.81% 18.31%
Market Capitalization MC in thousands $: Prior Month $79,666.872 $108,116.094

Trading Activity L: Prior Month .0288 .2762

Monthly Volatility GKPσ2 : First Day Transaction Month
GKPσ2 = [ln(Ot/Ct-1)]2 + .5[ln(Ht/Lt)]2 - .31[ln(Ct/Ot)]2
Term 1: [ln(Ot/Ct-1)]2 .0043 0
Term 2: .5[ln(Ht/Lt)]2 .0068 .0249
Term 3: .31[ln(Ct/Ot)]2 -.00003 -.0007
GKPσ2: Monthly Variance .0111 .0242
GKPσ Annualized Monthly Standard Deviation
:
.3649 .5389

percentage discounts. These discounts were applied to the Our discounted price for GGNS’s restricted stock was
prior month’s average share price for CRN and GGNS, (see $7.513 using the Bid-Ask spread model which was based on
Table VI) to calculate the discounted price for CRN’s and an average share price of $8.438 and an estimated discount
GGNS’s restricted stock. of 10.96%.27 The Longstaff discounted price for GGNS’s
Our discounted price for CRN’s restricted stock using the restricted stock was $1.958; based upon the average share
Bid-Ask spread model was $10.14. This was based on an price of $8.438, and the 76.79% discount.28 The discounted
average share price of $11.813, and the estimated 14.14% price reported by FMV was $6.8924, which implies a
discount.25 The Longstaff discounted price for CRN’s 18.31% discount from $8.438, the average transaction price
restricted stock was $6.10 which was based on the average in the month prior to the transaction date.
share price of $11.813 and an estimated 48.28% discount.26 While the Bid-Ask spread model produced better discounts
FMV reported a discounted price of $9.00, which implies a than the Longstaff equation, the model’s estimated discounts
23.81% discount from $11.813, its previous month’s average were slightly different than the reported FMV discounts.
transaction price.

25
$11.813 (1 - .1414) = $10.14. 27
$8.438 (1 - .1096) = $7.513.
                     
26
$11.813 (1 - .4828) = $6.10. 28
$8.438 (1 - .7679) = $1.958.           
          
Chipalkatti et al. – Estimating the Marketability Discounts 13

Table VII. Restricted Stock Discount Estimates

Calculated via Bid Ask Spread Model and the Longstaff Estimating Equation.
Bid Ask Spread Model: α0 + α1 ln(MC) + α2 ln(L) + α3 ln(GKPσ2)
α0 α1 α2 α3
3.6909 -.3737 -.1632 .2921
CRN 3.6909 -.3737(ln($79,666.872)) -.1632(ln(.0288)) .2921(ln(.0111))
3.6906 -4.2174 .5789 -1.3147
-1.2626
Model % Discount 14.14%
Reported FMV Discount 23.81%
GGNS 3.6909 -.3737(ln($108,116.094)) -.1632(ln(.2762)) .2921ln(.0242))
3.6906 -4.3315 .2100 -1.0870
-1.5176
Model % Discount 10.96%
Reported FMV Discount 18.31%

 σ 2 T   σ 2 T  σ2T  σ2T 

Longstaff Estimating Equation:  2   N  exp    1
 2   2  2π  8 

Term 1 Term 2
CRN
Longstaff % Discount 48.28% 1.2837 .1991
Reported FMV Discount 23.81%
GGNS
Longstaff % Discount 76.79% 1.4852 .2827
Reported FMV Discount 18.31%

We attribute part of this difference to the discrepancies we 2000, and employed accepted empirical models to estimate
observed between the FMV and CRSP share and volume data. the Bid-Ask spread and thereby the DLOM.
However, the differences between our discounted prices, The estimated DLOM using the Bid-Ask spread can range
and the observed discounted prices, were approximately one in value from 7% to 41% depending upon a firm’s attributes.
dollar per share, as compared to the Longstaff differences of We calibrated the performance of the two metrics using
approximately three and five dollars per share. Thus, we are actual restricted stock data and concluded that the Longstaff
confident that the model is useful for calculating discounts metric did not perform as well as the Bid-Ask spread based
for lack of marketability on firms that exhibit differences metric. We concluded that the percentage Bid-Ask spread
in size, trading activity, and volatility. We believe that the is a better proxy than the Longstaff metric for estimating the
predictive ability of the Bid-Ask model will improve with DLOM of illiquid stock. We believe that the predictive ability
the incorporation of financial statement data like profitability of the Bid-Ask model will improve with the incorporation of
and other relevant data like dividend yield. financial statement data like profitability and other relevant
data like dividend yield. Unfortunately such data were not
VI. Summary and Conclusion available for our sample period and this is a limitation of
this study. Since, both metrics can be estimated easily from
This paper’s goal was to improve the methodology used to firm data that are readily available, our results can be used
determine the discount for lack of marketability for closely to determine the DLOM for individual firms based on their
held firms, DLOM. We focused on two metrics as proxies for size, volatility, and trading activity. Additionally, the spread
the DLOM: the percentage Bid-Ask spread, and Longstaff’s based metric for DLOM is flexible enough to estimate a
lookback option. We used Over-the-Counter Bulletin Board, discount specific to an industry and a time period for which
OTCBB, Bid-Ask spread data for 1,850 firms that spanned a valuation is being conducted using appropriately tailored
the time period from December of 1995 until January of spread data.n  
14 Journal of Applied Finance – No. 2, 2012
Finnerty, J., 2002, "The Impact of Transfer Restrictions on Stock Prices,"
References Analysis Group/Economics Working Paper.
Bajaj, M., D. Denis, S. Ferris, and A. Sarinl, 2001, “Firm Value and
Marketability Discounts,” Journal of Corporate Law 27 (No. 1), 89- Kasper, L. 1997, Business Valuations: Advanced Topics, Westport, CT,
115. Quorum Publications.

Bilson, J., 2003, “An Assessment of Alternative Models of Financial Levine, L. and C. Luft, 2004, “Over The Counter Bulletin Board Exchange:
Market Volatility,” in Handbook of Risk, B. Warwick, Ed., New York, The Impact of Liquidity to Return, Volatility and Bid/Ask Spread,”
NY, John Wiley and Sons. Journal of Alternative Investments 7 (No. 3), 95-106.

Block, S., 2007, “The Liquidity Discount in Valuing Privately Owned Longstaff, F., 1995, “How Much Can Marketability Affect Security
Companies,” Journal of Applied Finance 17 (No. 2), 33-40. Values?” Journal of Finance 50 (No. 5), 1767-1774.

Bollen, N., T. Smith, and R. Whaley, 2004, “Modeling the Bid/Ask Spread: McMorrow, S., 2004, “New Environment of for the Lack of Marketability
Measuring the Inventory Holding Premium,” Journal of Financial Discount,” Business Appraisal Practice (Fall), 14-18.
Economics 72 (No. 1), 97-141.
Pratt, S., R. Reill, and R. Schweihs, 2000, Valuing a Business: The Analysis
Chacko, G., J.W. Jurek, and E. Stafford, 2006, “Pricing Liquidity: The and Appraisal of Closely-Held Companies, 4th Edition, New York,
Quantity Structure of Immediacy Prices,” Harvard Business School NY, McGraw-Hill.
Working Paper.
Robak, E., 2004, “The Current State of Marketability Discounts,” ACTEC
Chipalkatti, N., 2001, “Estimating the Discount for Lack of Marketability Journal 30 (No. 2), 137-145.
Using Proportional Bid-Ask Spreads,” Business Valuation Review 20
(No. 1), 3-10. Stockdale, J.J., 2008 “ A Test of DLOM Computational Models” Business
Valuation Review 27 (No. 3), 60-65.
Copeland, T. and D. Galai, 1983, “Information Effects of the Bid-Ask
Spread,” Journal of Finance 38 (No. 5), 1457-1469. Stoll, H. 2000, “Friction,” Journal of Finance 55 (No. 4), 1479-1514.

Damodaran, A., 2002, Investment Valuation, New York, NY, John Wiley Unites States Court of Appeals for the Sixth Circuit. 2001. Decision in re:
& Sons. Gross v. Commissioner, No. 99-2239/2257.

Damodaran, A., 2005, “Marketability and Value: Measuring the United States Tax Court. 2002. Tax Court Memo 2002-34 in re: Estate of
Illiquidity Discount” (July 30), Available at SSRN: http://ssrn.com/ Heck v. Commissioner.
abstract=841484.
United States Tax Court. 2003. Tax Court Memo 2003-258 in re: Lappo v.
Davidson, R. and J.G. MacKinnon, 1981, "Several Tests for Model Commissioner.
Specification in the Presence of Alternative Hypothesis,” Econometrica
49 (No. 3), 781-793. United States Tax Court. 2003. Decision in re: McCord v. Commissioner,
120 T.C. 358.
Duffy, R., 2009, “Exploring the New World of Quantitative DLOM
Analyses," Grant Thornton Business Valuation Monitor, April. United States Tax Court. 1995. Tax Court Memo 1995-255, June 12, in re:
Mandelbaum v. Commissioner.
Dyl, E. and G. Jiang, 2008, “Valuing Illiquid Common Stock,” Financial
Analysts Journal 64 (No.4), 40-47.