The Average Crap*

This is a posting on evaluating financial statements. Beneish M Score is a calculation to determine how likely it is that a statement is fraudulent. Generally the score is negative. As long as its is less than –1.78, the probability of fraud is low.

The score was developed by Messod Daniel Beneish, professor of accounting at the Kelley School of Busines at Indiana University.

The calculation for 3M for the year ended February 8, 2023, is

My conclusion, based on the score of –2.40, is that these statements are unlikely to contain fraud.

The score is the weighted total of eight indices.

The first index is day’s sales in receivables, a ratio of accounts receivable to sales divided by 365; it is a usual financial statement ratio. It shows on average how long it takes the company to collect its revenues. The index is the ratio of the current year to the prior one. If the company is recording sales that don’t exist, this should be greater than one.

The second index is gross margin index, that is, sales less cost of sales. Many financial statements show this as a subtotal on the income statement. The index is this year’s gross margin as a percent of sales to last year’s. Beneish believes that a declining margin is motivation for creative accounting.

The third is asset quality index. In some sense every asset is held to produce income. Each asset but cash and receivables has a life and should get written off over that life. Beneish identifies assets whose life and contribution to operations is clear. Those are current assets; plant, property, and equipment (PP&E); and securities, or quality assets. The asset quality ratio is (total_assets – quality_assets) ÷ total_assets.

Fourth is the sales growth index. That is this year’s sales to last year’s. Beneish believes that increasing sales may motivate management to overstate income in years when income does not meet expectations.

Fifth is depreciation index, the ratio of the percentage of PP&E written off this year to last. Overstating the value of PP&E and understating the depreciation charge are both techniques of managing income that are not generally accepted.

Selling, general, and administrative costs (SG&A) are the source of the next index. This year’s SG&A as a percent of sales is divided by last year’s. Subtract this index.

The leverage index, item 7, is the ratio of current liabilities plus long-term debt to total assets. It is a current-year statistic only. Beneish believes that highly leveraged firms are particularly likely to overstate income. Subtract this index from the score.

The last index is accruals to assets. That is (Cash provided by operations – Operating income)/Total Assets. Beneish’s observation is that misstatements rarely affect cash and therefore contribute to the difference between income and cash flow.

I wrote a workspace to perform this calculation. You can find it here. Fill in the array Beneish input. The function bi_Beneish_M will return the score and bi_Beneish_M_report will show the calculation of the score.

I found many websites explaining the Beneish M score and how to calculate it. Many noted that Enron scored in the fraud likely range but no one paid attention. Few websites actually offered a set of financial statements and calculated the M score.

One website that did calculate the M score got it wrong. The assets quality index (which measures the assets without quality [unquality being not a word.—KD]) was calculated with the quality assets as the numerator of the ratio. Their data was also flawed, as quality assets were greater than total assets.

~This all leads to my tentative conclusion: the Beneish M score, while it has demonstrated the ability to predict fraud, is too complicated. Beneish himself acknowledged that his eponymous score has too many false positives that require a great deal of digging to resolve.

My old laptop computer died, so I went and bought a new one. It arrived almost in time for my birthday, so the time I’ve spent setting it up was my birthday present.

Or rather I tried to convince myself that I was enjoying the challenge.

I have many happy memories of new computers with new or updated operating systems and discovering the innovative features of my new machine.

The last time I found that was with a Windows 3.1 computer. I had gotten my feet wet with a graphical interface, as one of my clients used Macs.

I was more than ready for running multiple programs at the same time and jumping from one to another. At that time I had a client who still posted his books by hand. Each time I went to see him, he presented me with a trial balance handwritten on ledger paper, and I dutifully copied everything into a file on my hard drive.

He in fact needed Windows more than I did. He had two desks, both covered with papers and open ledgers. He, like me, needed to refer to many sources as his work progressed; hence the second desk.

Unpacking my new box, I expected a voyage of discovery as I got the thing running, completely ignoring every version of Windows since version 3.1. Microsoft confuses change with innovation and thus changes the interface with each new version with little or no actual additional (or improved; often the opposite—KD) function. My voyage crashed on the rocks as the Windows logo opened on the laptop.

I had planned to make this a dual-boot machine, Linux or Windows. Step one was make a Windows recovery disk, and step two was install some basic applications, with Emacs and Open Office at the top of the list. I struggled and finally got to installing Linux.

Linux refused to partition the hard drive, which apparently Windows encrypts.. I got to a website that would “explain”—they fondly suppose—how to deal with the encryption. I gave up. I’m not a Windows fan anyway.

I still hadn’t given up hope of encountering some innovative features. I dove into the Linux install process. It didn’t take long to have a functioning laptop again. It was something of a relief that I had an administrator login and a user login, something still absent from Windows.

I started using Linux in the early 90s and have not looked back. Many years have brought a better install process, so the tedious things like partitioning and formatting the disk are done for you. Every year brings some new application or feature, and I’ve adopted many of them. All require installation and configuration. So while I quickly got back a working computer, I’ve spent the past two days installing the applications I wanted.

In terms of discovery, I discovered that Window Maker is spelled wmaker (Ugh—KD). We won’t discuss how long that discovery took or how many dead ends I went up in the discovery process.

I guess in some perverse way I still miss Windows 3.1. (It’s hardly perverse if it’s logical and reasonable.—KD)

Continue reading →

This post is a follow up to *Crunching Numbers in APL* available for kindle at Amazon.

Undaunted by my last foray into online analytical processing (OLAP), I sat down to code what I thought was OLAP. Workspace more_olap is here.

Should you load this workspace, you will find the following variables:

A fact in this exercise is the line item and amount columns from irdb.

irdb is made up of these columns:

      utl∆numberedArray ⊃ irdb[1;]
[001] Category code
[002] Category
[003] Subcategory code
[004] Subcategory
[005] Line item
[006] Fiscal year
[007] Amount (millions of dollars)

The function olap∆buildFacts loads the irdb data and produces all of these variables. You should also consider olap∆buildVars, which will produce variables to use as indices of facts.

      olap∆buildVars cats subs yrs
      )vars cat_
cat_a   cat_e   cat_l
      )vars sub_
sub_b       sub_cs  sub_da  sub_dfb     sub_dl      sub_i       sub_lo  sub_nn
sub_o       sub_oa  sub_oe  sub_ol      sub_orcv    sub_rcv     sub_re  sub_s
sub_sol
      )vars yr_
yr_2009     yr_2010

We now have some variables to use as indices of our fact cube and can look at our cube:

      olap∆combineFacts facts[cat_e;yr_2010]
 Paid-in capital                      11492
 Retained earnings                    28793
 Accumulated other comprehensive loss ¯3043

      ⍝ Or
      +/(olap∆combineFacts facts[cat_e;yr_2010])[;2]
37242

I still haven’t concluded that it’s easier than this:

SELECT Line_item, Amount from irdb where Category_code = 'e'
and Fiscal_year = 2010;

But I’m biased. While I’ve been writing APL code longer, I’ve spent more time writing SQL.

This is a simple exercise with simple data. It demonstrates what a data cube might look like and how to simplify slicing and dicing the data. There is no generalized code in this workspace, and therefore I must write a whole new workspace when I want to analyze a new data set.

I’m reminded of the weeks I spent designing a gross margin reporting system for a manufacturer. This company had several product lines and three departments. It had detailed time reports from the factory floor, so that I knew how much labor cost was incurred by product line and by department. It had a perpetual inventory system, so that I knew what material had been drawn from raw material inventory and production counts for each department. This allowed me to construct a model of the manufacturing process and estimates of costs incurred through each step in that process.

I’d like to get my hands on that long-lost data and see if a data cube would simplify anything.

OLAP stands for OnLine Analytical Processing. This post describes why
I’m not ready to write about OLAP.

I’ve been reading various web pages about OLAP and have reached two
conclusions. First, the demand for OLAP is driven by SQL commands’
complexity, which can arise when the programmer is querying complex
databases. Nontechnical users stumble badly in that environment, and
even correct queries can take too long to execute.

Second, the processing is done on a new non-SQL database designed to
make querying easier and processing time faster. Generally, this means
an underlying SQL database is kept to record transactions, and an OLAP
database is updated periodically from the SQL database. The OLAP
database usually is an array of facts determined by the SQL queries.
Each dimension of the array is a fact attribute for which aggregate data
may be sought.

I’ve been struggling to a third conclusion, that I can reach a better
understanding of the data by investigating why and how the data was
compiled than by constructing complicated SQL queries.

I found an open-source OLAP application, Cubes, written in
python([[http://cubes.databrewery.org/]]). I downloaded the application
and started on the tutorial. Step one, called Hello World, constructs a
data cube from balance sheets of the International Bank for
Reconstruction and Development for the years 2010 and 2011.

I was planning to code an OLAP database in APL, so rather than following
the tutorial, I just loaded the supplied csv file into APL.

      ibrd←import∆file 'Downloads/IBRD_Balance_Sheet__FY2010.csv'
      ⍴ibrd
63 7

That didn’t look like a lot of data, so I displayed some of it:

      ibrd[⍳5;]
 Category Code Category Subcategory Code Subcategory    Line Item                         Fiscal Year Amount (US$, Millions) 
 a             Assets   dfb              Due from Banks Unrestricted currencies                  2010                   1581 
 a             Assets   dfb              Due from Banks Unrestricted currencies                  2009                   2380 
 a             Assets   dfb              Due from Banks Currencies subject to restriction        2010                    222 
 a             Assets   dfb              Due from Banks Currencies subject to restriction        2009                    664 

My seven columns:

1. Category code
2. Category
3. Subcategory code
4. Subcategory
5. Description (called line item above)
6. Fiscal year
7. Amount (in $US millions)

I concluded that our cube should have two dimensions, description and
year. Each fact (cell in the array) should be made up of a description
and an amount. The category Descriptions is a hierarchy of category,
subcategory and item.

With that in mind I dived off the cliff and started writing APL queries.
Question one always is whether the debits equal the credits, or in this
case whether total assets equal total liabilities plus total equities.

I needed to confirm that both years and amounts were in fact numbers:

      utl∆numberp ¨ 1 0↓ibrd[⍳10;6 7]
1 1
1 1
1 1
1 1
1 1
1 1
1 1
1 1
1 1

Yes.
I knew I’d get tired of the column heads, so I copied the array without
them.

      db←1 0 ↓ibrd

I also determined the universe of category codes to simplify my next
query.

      db[;1]
 a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a l l l l l l l l l l l l l l l l l l l l l l e e e e e e e e 

I concluded that *a* means assets, *l* means liabilities, and *e* means
equity. My queries:

      +/(∊db[;1]='a')/db[;7]
558430
      +/(∊db[;1]='l')/db[;7]
480838
      +/(∊db[;1]='e')/db[;7]
77592
      480838 + 77592
558430

I started planning my next query and was curious: what descriptions
describe each fact?

      ⍞←⎕tc[3] utl∆join (db[;6]=2010)/db[;5]
Unrestricted currencies
Currencies subject to restriction
Trading
Securities purchased under resale agreements
Nonnegotiable, nonintrest-bearing demand obligations on account of subscribed capital
Investments
Client operations
Borrowings
Other
Receivables to maintain value of currency holdings on account of subscribed capital
Receivables from investment securities traded
Accrued income on loans
Net loans outstanding
Assets under retirement benefit plans
Premises and equipment (net)
Miscellaneous
All
Securities Sold under Repurchase Agreements, Securities Lent under Securities Lending Agreements, and Payable for Cash Collateral Received
Investments
Client Operations
Borrowings
Other
Payable to Maintain Value of Currency Holdings on Account of Subscribed Capital
Payable for investment securities purchased
Accrued charges on borrowings
Liabilities under retirement benefit plans
Accounts payable and misc liabilities
Paid-in capital
Deferred Amounts to Maintain Value of Currency Holdings
Retained Earnings
Accumulated Other Comprehensive Loss

I stopped.

I am not a bank accountant, so I must open the accounting rulebook and
read it cover to cover before I try to read banks’ financial statements.

Let’s consider a few facts. Accumulated other comprehensive loss is an
equity account that accumulates unrealized gains and losses. What they
might be is in the financial statements. We have just one page.

The subcategories make matters worse.

       ⍞←⎕tc[3] utl∆join (db[;6]=2010)/db[;4]
Due from Banks
Due from Banks
Investments
Securities
Nonnegotiable
Derivative Assets
Derivative Assets
Derivative Assets
Derivative Assets
Receivables
Other Receivables
Other Receivables
Loans Outstanding
Other Assets
Other Assets
Other Assets
Borrowings
Sold or Lent
Derivative Liabilities
Derivative Liabilities
Derivative Liabilities
Derivative Liabilities
Other
Other Liabilities
Other Liabilities
Other Liabilities
Other Liabilities
Capital Stock
Deferred Amounts
Retained Earnings
Other

When the Financial Accounting Standards Board issued guidance on
derivatives, I passed. I could do better in Atlantic City than with
derivatives, and so how to account for them was irrelevant. It certainly
seemed probable that these liabilities give rise to some of the
accumulated other comprehensive losses.

So after I complete my study of the accounting rules, I need to digest
the footnotes to the financial statements to get some understanding of
four different kinds of derivatives.

I have yet to extract an analysis of the facts I have. And that’s why
I’m not writing about OLAP…yet.
Continue reading →

When I took Accounting 1, I dreamed ledgers. In those days most medium to small companies kept their books by hand and stored them in a fireproof safe. One of the requirements of the course was a complete set of books consisting of financial reports for a fiscal period. It was all done by hand on ledger paper. I did manage to do some of it at work, where I could use a claculator [sic].

The dreams helped internalize accounting, and to this day if I have a difficult accounting problem, I’ll start with ledger paper and lay out my solution. At some point I’ll see the solution and then complete my work in APL.

What this means is that I am not afraid of long columns of figures nor of large arrays. I can extract insights just by examining the reports and doing some simple arithmetic.

Most people need something more, and a well-designed graph is always helpful. Accordingly, this post describes how to produce a graph in GNU APL and how to dress it up for company.

Graphing is not part of the ISO standard, but many APL interpreters provide a graphing function. In GNU APL it is ⎕plot. The syntax is attributes ⎕plot data. The handle return by ⎕plot is an integer that identifies the graph. Close the window with ⎕plot handle.

The data is what will be plotted. For a vector of real numbers, the data point will be positioned along the Y axis, and its position along the X axis by its position in the vector. Plotting according to pairs of data is done using complex numbers, which are the sum of a real number and an imaginary number. Thus, for a two-column array we could produce a graph by converting each row to a complex number:

      ⎕plot a_b[;1] + 0j1 × a_b[;2]

Dressing graphs up for company requires setting attributes. ⎕plot ” will give you a list of attributes.

I got into a heated discussion recently about U.S. tax policy, which led to the graph we’re about to construct.

I downloaded my data from BEA.gov and imported it into APL:

      income_expense_raw←import∆file '/home/dalyw/AverageCrap/Research/Federal_I_E.csv'
      gdp_raw←import∆file '/home/dalyw/AverageCrap/Research/GDP_2017Q1_2022Q4.csv'

For our purposes all we want is gross domestic product, line 6 in gdp_raw, and federal receipts, line 6 in income_expense_raw.

      gdp←gdp_raw[6;2↓⍳25]
      fed_receipts←income_expense_raw[6;2↓⍳25]

We want to plot both statistics against an actual timeline so that the X axis labels show quarterly increments.

      SPQ←×/91 24 60 60		⍝ Seconds in one quarter
      q1←⎕fio.secs_epoch 2017 2 15
      time←q1 + SPQ × ¯1 + ⍳23

We build the attribute array:

      ⍝ Set the GDP line color to blue
      att_gdp_tax.line_color_1←'#0000FF'
      ⍝ Set the tax line to green
      att_gdp_tax.line_color_2←'#00FF00'
      ⍝ Set the legend to identify both lines
      att_gdp_tax.legend_name_1←'Gross Domestic Product'
      att_gdp_tax.legend_name_2←'Tax receipts'
      ⍝ Position the legend away from the two lines
      att_gdp_tax.legend_X←50
      att_gdp_tax.legend_Y←200
      ⍝ Set the caption to identify the graph
      att_gdp_tax.caption←'US GDP and Tax Receipts in Billions'
      ⍝ Set the format of the X-axis labels to show year and quarter
      att_gdp_tax.format_X←'%yQ%Q'

Now we can call ⎕plot:

      att_gdp_tax ⎕plot (time + 0j1 × gdp),[0.1] time + 0j1 × fed_receipts

Quod erat demonstrandum.

This is post three of my Crunching Numbers in APL series. I’m
returning to my database of the top twenty stocks in the
Standard and Poor’s 500.

      sp20[;1 2 3 4 7]
 Symbol Name                        Price    Div $    FCF 
 WMT    Walmart                    142.09     2.28 ¯10929 
 AMZN   Amazon                      95.82     0     ¯1112 
 AAPL   Apple                      149.4      0.92  ¯2343 
 CVS    CVS Health                  86.04     2.42   2832 
 UNH    UnitedHealth Group         488.17     6.6    8651 
 XOM    Exxon Mobil                110.74     3.64  28024 
 BRK-B  Berkshire Hathaway         300.69     0         0 
 GOOG   Alphabet                    91.07     0         0 
 MCK    McKesson                   360.33     2.16      0
 ABC    AmerisourceBergen          159.5      1.94      0 
 COST   Costco Wholesale           493.14     3.6       0 
 CI     Cigna                      295.65     4.92      0 
 T      AT&T                        19.25     1.11      0 
 MSFT   Microsoft                  254.77     2.72      0 
 CAH    Cardinal Health             77.7      1.98      0 
 CVX    Chevron                    161.93     6.04      0 
 HD     Home Depot                 299.31     8.36      0 
 WBA    Walgreens Boots Alliance    36.21     1.92      0 
 MPC    Marathon Petroleum         125.52     3         0 
 ELV    Elevance Health            486.12     5.92      0 
 KR     Kroger                      43.91     1.04      0 
 F      Ford Motor                  12.07     0.6       0 
VZ     Verizon Communications      38.53     2.61      0

You’ll note I’ve added a column with some data. FCF is Free Cash
Flow. I’m using my own definition. I hope that it will act as
sieve to highlight stocks which deserve a closer look.

First I’d like to discuss databases. Chapter 11 of Crunching
Numbers in APL applies the principles of database design to APL
variables. We’re not to that point. We’re still in discovery
mode.

I set up a workspace for my research into companies that do not
pay dividends it includes the table sp20 shown above. As I’ve
done calculations I’ve tried to save those calculations and the
workspace as I played with Free Cash Flow. Here is where I
stand:

      )vars
aapl_free_cash      amzn_free_cash  cvs_free_cash   date∆US
date∆cal            date∆dates      date∆delim      date∆time∆M
date∆time∆delim     date∆time∆utce  date∆tz         final_vym
free_cash           g_data          g_return        goog
goog_covar          goog_free_cash  goog_hist       goog_variance
s_data              s_return        sp20            sp500
sp500_df            tmp             unh_free_cash   v_divs
v_lillian           vd_lillian      voo             vym
vym2                vym_div         vym_hist        wmt_free_cash
xom_free_cash

All the variables that begin date∆ belong the to
the date workspace in library 3 DALY and I’ll ignore them.

The variables sp500..., goog...,
and vym... were used for last week’s post on zero
dividend. The variables that end free_cash are for
today’s column.

APL’s workspace concept allows us this luxury. As I explore a
subject I can save my work in variables. They just exist and
don’t get in the way once I move on to something else.

For this project I created the variable free_cash and the
function calc_free_cash.

      free_cash
 Cash from operations       0 
 Interest                   0 
                      ------- 
 Adj Cash from ops          0 
 Capital exp                0 
 Dividends paid             0 
 Debt serv                  0 
 Stock repurchases          0 
                       ------ 
 Free cash                  0 
                      ======= 
 Debt service                 
 Interest                   0 
 Debt repayment             0 
 other                      0 
                      ------- 
                            0 

     ∇rs←calc_free_cash fc                                              
[001] ⍝ Function calculate free cash flow from a free_cash 
[002] ⍝ workpaper and returns a free_cash workpaper with 
[003] ⍝ those results.               
[004] rs←fc                                                             
[005] rs[4;2]←+/rs[1 2;2]                                               
[006] rs[13;2]←-rs[2;2]                                                 
[007] rs[7;2]←rs[17;2]←+/rs[13 14 15;2]                                 
[008] rs[10;2]←+/rs[4 5 6 7 8;2]                                        

My idea is a measure of the cash available from the operations
that can be used to grow the company. Today’s Wall Street
Journal has an article on evaluating companies that pay
dividends. It recommends ignoring the amount of the dividend and
instead focusing on cash flow.

The financial statements have five basic statements:

• Balance Sheet
• Income Statement
• Comprehensive Income
• Stockholder’s Equity
• Cash flow

My free cash flow calculation pulls amounts from that last
statement, Cash flow. Its worth looking the statement as whole.

First it reconciles net income to cash from operations. That
reconciliation includes items used to calculate net income which
do not use or provide cash, depreciation for example. The
reconciliation also includes changes in working capital that
require or provide cash.

Second it shows investment activity. I get my capital
expenditures from this section. I know that the company must
replace plant, property and equipment as it wears out. I use
this line as an estimate for future operations.

Third it shows financing activity, debt and equity
transactions. Here I find the amount of dividends paid and stock
repurchased. I calculate debt service as interest (from the
income statement) plus debt repayment for this section of the
cash flow statement.

The decision to finance the company through debt requires
consideration of the payment of interest and the retirement of
principle. I recognize this by adding interest to cash from
operations, and including it in debt service.

Many things may be said about this approach. It is
too simple. A thorough reading the financial statements and
Management’s Discussion and Analysis of Financial Condition and
Results of Operations might yield better estimates. In fact
those estimates my be buried in the 10-K somewhere.

This method is quick and dirty but I like it.

MCK, McKesson, is next on my list. I found its 10-K for the year
ended March 31, 2022 at www.sec.gov and its statement of cash
flow on page 74. Here is how I calculate free cash flow.

      mck_free_cash←free_cash
      mck_free_cash[1;2]←4434
      ⍝ This from the bottom of the cash flow statement
      mck_free_cash[2;2]←186	
      ⍝ The total of property, plant and equipment and software
      mck_free_cash[5;2]←¯388 + ¯147 
      mck_free_cash[6;2]←¯277
      ⍝ Repayment of long-term debt and debt extinguishments
      mck_free_cash[12;2]←¯1648 + ¯184 
      mck_free_cash[8;2]←¯3516

      ⍞←mck_free_cash←calc_free_cash mck_free_cash
 Cash from operations    4434 
 Interest                 186 
                      ------- 
 Adj Cash from ops       4620 
 Capital exp             ¯535 
 Dividends paid          ¯277 
 Debt serv              ¯2018 
 Stock repurchases      ¯3516 
                       ------ 
 Free cash              ¯1726 
                      ======= 
 Debt service           ¯1832 
 Interest                ¯186 
 Debt repayment             0 
 other                      0 
                      ------- 
                        ¯2018 

Now I’ll update my database and cross McKesson of my list.

      sp20[10;]
 MCK McKesson 360.33 2.16 21.79 263966 0 
      sp20[10;7]←¯1726
      sp20[;1 2 3 4 7]
 Symbol Name                        Price    Div $    FCF 
 WMT    Walmart                    142.09     2.28 ¯10929 
 AMZN   Amazon                      95.82     0     ¯1112 
 AAPL   Apple                      149.4      0.92  ¯2343 
 CVS    CVS Health                  86.04     2.42   2832 
 UNH    UnitedHealth Group         488.17     6.6    8651 
 XOM    Exxon Mobil                110.74     3.64  28024 
 BRK-B  Berkshire Hathaway         300.69     0         0 
 GOOG   Alphabet                    91.07     0         0 
 MCK    McKesson                   360.33     2.16  ¯1726 
 ABC    AmerisourceBergen          159.5      1.94      0 
 COST   Costco Wholesale           493.14     3.6       0 
 CI     Cigna                      295.65     4.92      0 
 T      AT&T                        19.25     1.11      0 
 MSFT   Microsoft                  254.77     2.72      0 
 CAH    Cardinal Health             77.7      1.98      0 
 CVX    Chevron                    161.93     6.04      0 
 HD     Home Depot                 299.31     8.36      0 
 WBA    Walgreens Boots Alliance    36.21     1.92      0 
 MPC    Marathon Petroleum         125.52     3         0 
 ELV    Elevance Health            486.12     5.92      0 
 KR     Kroger                      43.91     1.04      0 
 F      Ford Motor                  12.07     0.6       0 
 VZ     Verizon Communications      38.53     2.61      0 
 JPM    JPMorgan Chase             139.67     4         0 
 GM     General Motors              39.25     0.36      0 

This is the second in a series of posts about using APL to crunch numbers. It starts with a database of sorts in APL.

I want to use present-value equations to assess various stocks. To do so I need specific stocks to investigate. I built a database of the top 20 stocks from the Fortune 500. I did have to do this by hand.

finance.yahoo.com is a good source for information about publicly traded stocks, bonds, and mutual funds. I got a list of the Fortune 500 and went to Yahoo. Here is what I compiled:

 Symbol Name                        Price    Div $    EPS  Total Rev   
 WMT    Walmart                    142.09     2.28   4.27 611289       
 AMZN   Amazon                      95.82     0     ¯0.28 513983       
 AAPL   Apple                      149.4      0.92   5.9  395328       
 CVS    CVS Health                  86.04     2.42   3.14 322467       
 UNH    UnitedHealth Group         488.17     6.6   21.17 322132       
 XOM    Exxon Mobil                110.74     3.64  13.26 398675       
 BRK-B  Berkshire Hathaway         300.69     0     ¯0.97 345636       
 GOOG   Alphabet                    91.07     0      4.54 282836       
 MCK    McKesson                   360.33     2.16  21.79 263966       
 ABC    AmerisourceBergen          159.5      1.94   8.25 238587       
 COST   Costco Wholesale           493.14     3.6   13.23 226954       
 CI     Cigna                      295.65     4.92  21.29 180642       
 T      AT&T                        19.25     1.11  ¯1.1  120741       
 MSFT   Microsoft                  254.77     2.72   9    198270       
 CAH    Cardinal Health             77.7      1.98  ¯4.56 181364       
 CVX    Chevron                    161.93     6.04  18.28 235717       
 HD     Home Depot                 299.31     8.36  16.68 157403       
 WBA    Walgreens Boots Alliance    36.21     1.92  ¯3.43 132703       
 MPC    Marathon Petroleum         125.52     3     27.98 177453       
 ELV    Elevance Health            486.12     5.92  24.81 156595       
 KR     Kroger                      43.91     1.04   3.18 137888       
 F      Ford Motor                  12.07     0.6   ¯0.49 158057       
 VZ     Verizon Communications      38.53     2.61   5.06 136835       
 JPM    JPMorgan Chase             139.67     4     12.1  128641       
 GM     General Motors              39.25     0.36   6.09 156735

All of this data and a lot more is available issue by issue at Yahoo. I
chose this data to help determine which issues are worth further
investigation. The statistic I was most interested in was the dividend
yield, column four divided by column three.

Financial theory proposes that a stock’s price is dividend ÷ (yield –
growth). Yield is expected to be your return on the investment, and
growth is the rate at which the dividend is expected to grow.

So before I could work on estimating growth and/or yield, I needed to address Alphabet, which has never paid a dividend but whose stock is quite valuable.

When I learned financial theory, I decided that companies that paid no dividends had no value; just do the arithmetic with any yield or growth assumption. Today’s post evaluates the zero-dividend strategy.

I needed a baseline to compare to Alphabet’s performance. I choose
Vanguard high-yield dividend fund (VYM). It is a mutual fund generally made up of dividend-paying stocks—high dividend if you believe the title. Vanguard is well known for mutual funds that provide similar returns to the market as a whole.

I went to Yahoo and downloaded the dividends paid by VYM over the 10 years ended 12/31/2022.

      vym_div←date∆US import∆file∆withDates '/home/dalyw/Downloads/VYM.csv'
      ⍴vym_div
41 2
      vym_div[⍳15;]
 Date            Dividends 
 2013 3 22           0.361 
 2013 6 24           0.419 
 2013 9 23           0.437 
 2013 12 20          0.532 
 2014 3 24           0.401 
 2014 6 23           0.476 
 2014 9 22           0.469 
 2014 12 18          0.562 
 2015 3 23           0.462 
 2015 6 26           0.56  
 2015 9 23           0.528 
 2015 12 21          0.599 
 2016 3 15           0.478 
 2016 6 21           0.578

I also looked up the opening price of VYM on 1/1/2013—$45.89—and 1/1/2023—$108.21. I could now produce a date flow.

The workspace ‘5 DALY/fin’ distributed with gnu-apl. It has present and future value functions for date flows. A date flow is an ordered
collection of date–amount pairs. Each date is in Lillian format, that
is, the number of days from October 15, 1582, the first day of the
Gregorian calendar.

I took this data and assembled a date flow that assumed the purchase of VYM on 1/1/2012 and its sale on 12/31/2022:

      fin∆df∆show vym
2013/01/01         (45.89)
2013/03/22           0.36 
2013/06/24           0.42 
2013/09/23           0.44 
2013/12/20           0.53 
2014/03/24           0.40 
2014/06/23           0.48 
2014/09/22           0.47 
2014/12/18           0.56 
2015/03/23           0.46 
2015/06/26           0.56 
2015/09/23           0.53 
2015/12/21           0.60 
2016/03/15           0.48 
2016/06/21           0.58 
2016/09/13           0.48 
2016/12/22           0.67 
2017/03/22           0.56 
2017/06/23           0.60 
2017/09/20           0.60 
2017/12/21           0.64 
2018/03/26           0.61 
2018/06/22           0.63 
2018/09/26           0.67 
2018/12/24           0.74 
2019/03/25           0.65 
2019/06/17           0.62 
2019/09/24           0.79 
2019/12/23           0.78 
2020/03/10           0.55 
2020/06/22           0.84 
2020/09/21           0.70 
2020/12/21           0.81 
2021/03/22           0.66 
2021/06/21           0.75 
2021/09/20           0.75 
2021/12/20           0.94 
2022/03/21           0.66 
2022/06/21           0.85 
2022/09/19           0.77 
2022/12/19           0.97 
2023/01/01         108.21

I computed an internal rate of return using fin∆df∆irr as 12.04%.

I went back to Yahoo for the opening price of Alphabet stock on 1/1/2013 and on 1/1/2023 and calculated the growth of the stock’s value.

      goog←((date∆lillian 2012 1 1) ¯16.26 ) fin∆df∆add (date∆lillian 2023 1 1) 89.86
      fin∆df∆irr goog .1
0.1552653407

Here I set up a date flow that assumes the purchase of stock on 1/1/2013 and its sale on 1/1/2023 and then calculate an annual return of 15.527%.

This certainly challenges my working hypothesis that a stock that pays no dividends is worthless. Had I bought Alphabet 10 years ago, I would have realized gains greater than the market, the holy grail of
investing.

I then went to
https://www.sec.gov/edgar/searchedgar/legacy/companysearch.html
Public companies are required to register with the SEC and file detailed reports on their operations. The annual report is 10_K, which for GOOG I opened.

Page 1 had the first yellow flag. There are two classes of stock, A and
C, registered with the SEC. I wondered about B. Page 2 is the table of
contents with hyper links to various sections. I went to the financial
statements and read the notes. Class A, 6015 shares, allow 1 vote per
share. Class B, 893 shares, allow 10 votes per share and are not
publicly traded. Class C, 6334 shares, have no voting rights. I read the equity footnote twice. It addresses the rights of the various shares and makes the case that they share equally on liquidation of the company. What happens should Alphabet declare a dividend was not clear.

In summary, the company is absolutely controlled by the holders of the class B shares who apparently are not interested in receiving dividends.

I looked at the rest of the financial statements and made my own
estimate of free cash flow. That is, the cash generated from operations that could be used to pay dividends or repurchase stock.

               In millions
 Cash from operations        91495 
 Interest                        0 
 Stock based awards          ¯9300 
                           ------- 
 Adj cash from operations    82195 
 Capital Expenditures       ¯31485 
 Debt Service                ¯1196 
 Dividends                       0 
 Stock purchases            ¯59296 
                           ------- 
 Free Cash                   ¯9782 
                           =======

Not an encouraging picture. Alphabet is using its accumulated cash to buy back its stock. That suggests that the company can find no better investment.

High-growth companies do not pay dividends because they need every dollar of cash to support their growth. Clearly, Alphabet is no longer in this category, and it’s high time it shared the lolly. The promise of that growth suggests that the company can get higher returns from its operations than the stockholders can by reinvesting in other stocks.

I don’t think I’m changing my approach to zero dollar dividends.

Well, it’s done. That is, I’ve published the book I’ve been working on for the past three years. I never thought it would take three years.

You can buy it for your Kindle. Search for its title Crunching Numbers in APL.

Too much of this book is how to code APL and not enough of how to use APL. It is an unavoidable sojourn if you want to use APL to crunch numbers. So I’m going to write a series of posts on how to use APL.

We’ll start with inflation.

I suffer from my Wharton education, and one of the sources of that suffering is inflation. I attended Wharton during the seventies, when inflation got out of control. Wharton historically stresses monetary economic policies over Keynesian, and Milton Friedman was banging the drum for better control of the money supply. Washington has never really wanted to control the money supply, although during the eighties it had to.

I downloaded the Federal Reserve data on the money supply. (https://www.federalreserve.gov/releases/h6/current/default.htm) and created this graph:

You’ll note the sharp increase in the money supply in January through June of 2020. This was the period when Congress enacted blowout bills that it believed would help people in economic distress because of the pandemic.

Money supply keeps growing to November of 2021, when the Federal Reserve executed policies to reduce the rate of inflation. It’s now February of 2023, and the rate of inflation has moderated but is still too high.

The curve shows a small decline as the Fed acted.

How did I construct this graph?

I downloaded the file from federalreserve.gov and imported it into APL. This gave me a table 774 lines by 30 columns. Line one was a description of each column. I picked column one, the year and month; column two, seasonally adjusted M1; and column three, seasonally adjusted M2. I now had a table 769 lines by 3 columns.

I made several graphs using ⎕plot. My first try was

I got:

I was alarmed. I had hoped I didn’t have to figure out the differences between M1 and M2, but when I graphed M2, I got what I expected:

I consulted the oracle internet. M1 is generally currency in circulation plus bank demand deposits (read checking accounts). M2 is M1 plus net time deposits (read savings accounts). I found that the Federal Reserve had changed the banking rules in April of 2020. Just showing the underlying data shows how banks responded:

      M1_M2[720+16 17 18;]  
2020-03  4261.9 15988.6  
2020-04  4779.8 17002.5  
2020-05 16232.9 17835.2

I dressed the M2 graph up for company with gnuplot.

I started to write this post as “User Friendly,” but after reading all sorts of blogs on that subject, I changed it.

Each of these blogs proposed a list of four or five features that describe user-friendly software. Here is my own summary:

1. Simple
2. Clean
3. Intuitive
4. Reliable

There were many words surrounding these four. In some blogs there appeared to be some meaning associated with those words. In many there was none. All seemed redundant. [Sort of like describing sympathetic as having sympathy—KD]

Simple is perhaps the most difficult. How often have you confronted a simple idea that you wanted to internalize and discovered as you wrestled with it exactly how complicated is was?

Clean follows simple. As I read, I kept finding clean described as simple. Part of me wants to make clean a complement to simple or perhaps combine them. A clean and simple interface. I can’t imagine either clean or simple without the other.

Reliable is a new concept. Does the software always do what it is supposed to do? Is it buggy?

Intuitive is last. Every blogger believed that a user-friendly interface will allow the user to know just by looking at the screen what to do next.

Merriam-Webster defines intuition as (1) “the power or faculty of attaining to direct knowledge or cognition without evident rational thought and inference,” (2) “immediate apprehension¹ or cognition” (https://www.merriam-webster.com/dictionary/intuition) I’m the wrong person for this idea. I won’t say that I don’t ever get flashes of insight; I do. I also know that those flashes come only after a struggle of rational thought and inference. Athene has not sprung full-grown and fully armed from my brow.

¹ Apprehension here is used in MW’s third sense: perception, comprehension. It has nothing to do with fear.—KD

As I struggled with the concept of user friendliness, I thought I ought to look again at Excel, in some minds the epitome of user friendliness. [What a crock. I could tell stories…—KD] So I rebooted my machine from Debian to Windows (always painful) and started Excel. Once I had a spreadsheet loaded and was contemplating how to test friendliness, it struck me. One of my bloggers had offered up MS Office as an example of user hostility. His issue was the ribbons that Microsoft implemented several years ago and the struggle their user base had adapting.

How often is change used to simulate innovation? How often has Microsoft labeled change as innovation? [Microsoft is a piker compared with the textbook publishers; I could name a few textbooks whose six editions all had the same material. Way to kill the aftermarket, guys.—KD] I went back to Debian.

I remember switching to Quattro Pro from Lotus 1-2-3. At first it was dollars and cents. Quattro Pro was less than one fifth the cost of Lotus 1-2-3. I remembered the joy I felt using Quattro Pro. It had more features, but was it user friendly?

I found an old backup of Quattro Pro [The triumph of the hoarder—KD] and copied it onto my hard drive. This, it turned out, was all the installation I needed. Debian provides DOSBox, a DOS emulator. I used it to start Quattro Pro. I tried building a simple spreadsheet to remind myself how it worked.

Things were turned around. If I wanted to copy, I first selected copy from the menu; typed in the upper left and lower right corners of the source block and typed <ENTER>. I then typed in the upper left corner of the destination and <ENTER> again. I’m used to highlighting the source block with my mouse, right-clicking for a menu, and selecting copy.

While the mouse and the GUI interface changed how software worked, the keyboard procedure was easy to understand and to use. The latest version of Excel has many enhancements, some of which speed up construction of a spreadsheet. It also has a lot more functions on its ribbons—if only I could remember which ribbon and what the pictures on the ribbon actually mean.

How long did it take you to understand what a pivot table is? Can you find it on a ribbon on the first try?

I left out one idea that my research uncovered: the principle of least astonishment. “The behavior [of the software] should not astonish or surprise users.” (https://en.wikipedia.org/wiki/Principle_of_least_astonishment)

In my use of APL I’ve been struck over and over again how easy it is to type in a line of code and have it do exactly what I thought it would. APL uses an old-fashioned teletype-like interface to do powerful things. Because of the simplicity and cleanliness of its design, I can do those powerful things.