Tag Archives: valuation

Quantitative Analysis and Multiple Linear Regression Modeling using StockCalc Valuation data

Quantitative Analysis and Multiple Linear Regression Modeling
using StockCalc Valuation data

September 2021

Executive Summary:

StockCalc is a fundamental valuation platform. We generate valuations each night for the 8000 equities trading on NYS, NAS, ASE, TSX and TSX-V using an algorithmic approach that has been historically correct 78% of the time.

To build upon our existing product, we created a series of multiple linear regression models. We included the 6 fundamental valuation data points we generate each night (described below) along with technical indicators (Moving Averages, RSI, MACD) and commodity prices.

Using the R programming language we built linear regression models for each stock each night to enable us to test the dependent variables of recent close price, price 13, 26 or 52 weeks later against the series of independent variables (our fundamental valuations, technical indicators and commodity prices)

Our goal was to determine if we could improve upon our current approach to calculating a weighted average valuation (“The Algo”) by using this MLR approach (“The MLR”). Our current approach is correct 78% of the time going out 12 months. That means our valuation is reached 78% of the time over the next 12 months (we refer to as %success) with an average 10.2% difference between close price and our valuation (we refer to as %difference).

From the Analysis …

Near dated regression models (using most recent Close price) were able to outperform The Algo. We looked at both % success rate and % difference (price versus valuation) when making this determination.

Of note: the % difference we refer to in this document shows what the results would be if we assume returns are locked in when price reaches our valuation. Part of the analysis then looked at additional upside if we used a trailing stop protocol to let winning stocks continue to run out for 52 weeks. Those results exceed 40% annual returns once we netted winners against losers (winners up 77%, losers down 20 %)

Stocks on average reach our valuation in 13 weeks implying we can be running these models multiple times each year to magnify the returns. We will be doing more work on this to better quantify results for multiple successful models per year.

Contents

Executive Summary:
Focused on valuation; empowered by analytical tools.
Background:
“The Algo”
Methodology aka “The MLR”
SQL
Back-testing:
Initial Results:
Improving the Results:
The difference between markets:
The finer details:
There is more upside here:
How can you use/access this data?
About the author
Appendix 1: Description of the 5 fundamental valuation models used in Stockcalc
Appendix 2: Independent variables available for the regression models

Figures and Tables

Table 1 Initial regression results – various indexes
Table 2 Extended Results – Capturing 10% or more difference
Table 3 Comparing pre/including Covid results

Figure 1 Regression set-up for 13, 26 and 52 week analysis
Figure 2 sample regression logic using ADBE:NAS

Focused on valuation; empowered by analytical tools.

StockCalc is a company focused on providing the platform and data to enable its clients to undertake Fundamental Valuation of publicly traded companies. The modules that have been created help those looking to analyze companies arrive at a more thorough understanding of the value of the equity in the company they are examining.

Background:

The Stockcalc database contains 8000 equities that are run through 5 fundamental valuation models each night. We also examine Analyst consensus data and will use it if it out-performs our analysis (This occurs 5-15% of the time) Our data and analysis is available for advisors and analysts on TD Ameritrade’s Veo open Access platform as well as Schwab’s Advisor platform. StockCalc is also available in the Globe and Mail platform in Canada.

We generate up to 6 valuation points for each company in the database each night.

DCF: Discounted Cash Flow
PComps: Price Based (PE, PB, PCF) comparables
OComps: Other comparables (for companies that are not profitable)
Mul: Multiples (A comparables using the companies’ historic data)
ABV: Adjusted Book value
Analyst: Consensus Analyst Target Value

A detailed description of each of these is available in Appendix 1

Each night once all the models have been run for the 8000 stocks in the database we generate a Weighted Average Valuation (WAV) for each stock using an algorithm that selects the various models based on a series of criteria (industry, historic accuracy, range) and weighting.

From 2013 to mid 2020 our WAV values have been correct 78% of the time for the S&P 500 list of companies looking out over the next 12 months (ie Close price moved beyond our WAV over the next 12 months 78% of the time). We wanted to see if we could improve on that by generating an MLR model for each company each day. Using an MLR model approach also allowed us to incorporate other independent variables such as technical indicators and commodity prices.

“The Algo”

Looking at the S&P 500 list The Algo modeling has been correct 78% of the time: We define that as the price reached our valuation (either to the up or down side) 78% of the time over the next 12 months. By comparison, analyst consensus data is correct on average 50% of the time (various studies)

From 2013-01-01 to 2020-06-30 (ie 52 weeks ago so we can test against next 52 weeks)

Our Dataset has 368,221 records of daily weighted average valuations (we call those our WAV’s)

# records where our WAV was > ClosePrice and next 52 Week high > WAV: 140,897
# records where our WAV was <= ClosePrice and next 52 Week low <= WAV: 146,845
(140897 + 146845) / 368221 = 78%

Notes: models were initially constructed first of the month until Aug 2018 when we started running them daily for the 8000 stocks in the database.

Methodology aka “The MLR”

The regression set-up:

Each night we run the fundamental models against the 8000 stocks in our database. We constructed code using the R programming language (https://www.r-project.org/about.html) to enable us to select the data and build the regression models. We used the same daily valuation data as mentioned in The Algo to ensure consistency. We then constructed a series of linear regression models using R. Each stock was regressed each day after 50 days of data was available (generally the models started early 2017 as we had monthly data until 2018)

We tested four (4) time frames with our modeling:

Recent close price: We programmatically built the regression models using all data from 2013-01-01 to the run date less 1 day then plugged the current days model values (Fundamentals like DCF, Comparables etc along with technical or commodity prices if selected by the regression equation) to generate what the recent close price should be based on the regression equation.

Price 13, 26 and 52 Weeks later: For the other 3 time frames we needed to go back 13, 26 or 52 weeks into the data to construct the regression models as our dependent variable was the actual price 13, 26 or 52 weeks later so we could not use their present day values. We then used the recent day fundamental, technical or commodity prices (as selected by the regression equation) to generate a next 13, 26 and 52 week price estimate.

It is very important to re-iterate for these three time frames we needed to go back 13, 26 or 52 weeks into the data to construct the regression models as our dependent variable was the actual price 13, 26 or 52 weeks later. This meant we would be using data from 2013-01-01 to a given date to generate the regression equation then plugging in the data from the date we were calculating which was 13, 26 or 52 weeks later. You will see the impact in the results tables later on. The following figure depicts the set-up:



Figure 1 Regression set-up for 13, 26 and 52 week analysis

SQL:

The initial results from the r programs were written to excel then imported into our SQL database for easier processing. Our SQL database contains all the analysis we undertake each night along with the financial, ratio and price data used to generate the valuations.

Determining Model Success:

To determine if our estimate calculations were correct we compared our estimate price to most recent close to first determine if we show the stock as under or over-valued. We then looked out 13, 26 or 52 weeks to see if the High (if undervalued) or Low (if Overvalued) went beyond our estimate price. If they did we considered the model successful for that stock, date and time frame.

In the table below we show these details for Adobe (ADBE :NAS) as of Dec 31, 2018 close.

There are four reference points: Close, 13, 26 and 52 which refer to when (in the past) the regression equation was constructed (using last nights close data back to 52 weeks ago) The data from the date we are running was then run though each of those regression equations to determine what we calculate the price would be Today, 13 , 26 and 52 weeks from now.



Figure 2 Sample regression logic using ADBE:NAS

Our valuation database contains analyses going back to January 1st, 2013. For each company in the S&P 500 we generated an MLR for each date up to June 30th 2020 (52 Weeks), Dec 30th (26 week), (Mar 31st 2021 (13 Week) and June 30th 2021 (Recent close)

Initial Results:

We started by selecting all stocks in a given ETF to proxy the index lists. We used the DIA (Dow 30), SPY (S&P 500), QQQ (Nasdaq 100), IWM (Russell 2000) and XIU (Canada – SP/TSX 60). In the r code we ran each stock each day once we had 50 days of valuation available (we had monthly valuations 2013-mid-2018 then daily thereafter).

From the table below you see the initial results: For example our Algo (WAV columns) for the SPY was correct 76.93 % of the time with an average spread of 7.62% between the close price and what our valuation calculation was. Looking across the columns you see the close price had the highest % success due to the smallest time frame. It also had the least spread between price and value. For the DIA and SPY The Algo outperformed the regressions built with 52 and 26 week old data but the 13 week old data was more correct than our Algo with higher % difference between price and value.

The r2 column shows the minimum value selected based on each equations r2 value. For each model each day we captured the r2 value and used it to help in the selection. We did not track r2 for WAV.

This was the base case run to understand how The Algo compared to the MLR.


Table 1 Initial regression results – various indexes

Improving the Results:

In the next step we started to select sub-groups of data to run though the regression models. In this example we selected all stocks that showed a 10-50% difference between price and our valuation (for the Close, 13, 26 and 52 week analysis).

In the table below you can see there are 2 things happening compared to the previous table. Our average % difference between price and value has jumped to between 12 and 25% and our % success has been reduced due to the higher spread between price and value.

Note how well the close price columns performed in this table. We see a 75 to almost 90% success rate using 1 day old data to generate the regression equations with the % difference ranging from 12 to 17%. (Again noting results shown assume returns are locked in when price reaches our valuation – more on this below)

The Algo was able to outperform on the XIU data here (S&P/TSX 60 stocks) is also of note.


Table 2 Extended Results – Capturing 10% or more difference

The difference between markets:

As you can also see from the tables above we show smaller % difference between price and value for the DOW (DIA) than the SPY and IWM. This inherently makes sense given the coverage and ownership the DOW 30 stocks have compared to for example the Russell 2000 list.

The finer details:

To eliminate the impact of Covid on the markets we also tested data from 2017-2019. You can see the numbers remained consistent with the 2018 markets dragging down overall results slightly.


Table 3 Comparing pre/including Covid results

The real value starts here:

The results shown above assume returns are locked in when price reaches our valuation. To fine tune this further we selected the S&P 500 list of Undervalued stocks to Dec 31 2019 (to eliminate Covid impact again) where our models were 20% or higher than close price. We had 40 stocks during that time period meeting this criteria (64 records including Stocks that met the criteria more than once).

Thirty-five of these stocks were successful (our valuation target was exceeded over next 12 months) for an average return of 77% when we let these stocks run out for 52 weeks. The 5 stocks that were not successful had an average return of -20% 52 weeks later for a combined return of 42.5%. A trailing stop program would easily capture this additional return.

In our database, our WAV valuation models on average are being reached within 13 weeks. This implies the returns we presented in Table 2 could be achieved four times each year. The net 13 week return column in Table 2 ranges from 10 to 15% therefore 40% to 60% returns less negative returns for those stocks not achieving valuation.

The next steps in this process will be analyzing the data further to construct a protocol that can select the superior model (WAV, Close, 13) as well as reassessing when price reaches our valuation to see if we would continue to hold the equity.

How can you use/access this data?

The ability to slice and separate the data is inherent in the r code and resulting SQL database. We are well positioned to assist you with this data and analysis. StockCalc has a powerful API that provides connections for funds and firms looking to access this data. For more information please reach out to Brian to discuss

About the author

Brian Donovan is the President of StockCalc and is a Chartered Business Valuator (CBV – CICBV Website). He obtained in Finance MBA in 1999 from Wilfrid Laurier University in Canada and worked in a variety of capacities from MyPlant.com in the dot com days to founder of StockCalc in 2013. Brian can be reached by email at brian.donovan@stockcalc.com

Appendix 1: Description of the 5 fundamental valuation models used in Stockcalc

Discounted Cash Flow (DCF)
Discounted Cash Flow (DCF) valuation is a cash flow model where cash flow projections are discounted back to the present to calculate value per share. DCF is a common valuation technique especially for companies undergoing irregular cash flows such as resource companies (mining, forestry, oil and gas) going though price cycles or smaller companies about to generate cash flow (junior exploration companies, junior pharma, technology firms…).

Price Comparables
The Price Comparables valuation is the result of valuing the company we are looking at on the basis of ratios from selected comparable companies: Price to Earnings, Price to Book, Price to Sales, Price to Cash Flow, Enterprise Value (EV) to EBITDA. Each of these ratios for the selected comparable companies are averaged and multiplied by the values for the company we are interested in to calculate a value per share for our selected company.

Other Comparables
We have included the Other Comparables as a way to value companies that cannot be valued using Earnings based ratios. This technique is very useful for companies still experiencing negative cash flows such as mining exploration firms. We use Cash/Share, Book Value/Share, MarketCap, 1 Year Return, NetPPE as the ratios here. Each of these ratios for the selected comparable companies are averaged and multiplied by the values for the company we are interested in to calculate a value per share for our selected company.

Multiples
Multiples are similar to Price comparables where we look at current or historic ratios for the company in question to assess what it should be worth today based on those historic ratios. We use the same 5 ratios as in the price comparables and value the company with its historic averages.

Adjusted Book Value
With Adjusted Book Value (ABV) we calculate the book value per share for the company based on its balance sheet and multiply that book value per share by its historical price to book ratio to calculate a value per share.

Analyst Consensus
If we have Analyst coverage for the company we use the consensus target price here.

Appendix 2: Independent variables available for the regression models

Valuation:
DCF, P_Comps, O_Comps, Multiples, Adjusted Book, Analyst Consensus

Technical:
30 and 200 Day moving averages, MACD, RSI

Commodity:
Natural Gas, WTI, Brent, Gasoline, Propane
Aluminum, Copper, Nickel, Steel, Zinc, Iron Ore, Coal, Cobalt, Lead, Moly, Tin
Gold, Silver, Platinum, Palladium,
Lumber, Corn

Dow 30 Stocks Price vs Valuation – April 2021

Are the markets becoming overvalued? Using StockCalc.com we ran the DOW 30 to look at Price versus valuation and currently we see 18 DOW 30 Stocks at or above fair value.

At StockCalc each day we run fundamental valuation models on 8000 Stocks (Canada, USA – TSE, NYS, NAS). I wanted to share this graph to show the valuations we have for the list of DOW 30 stocks currently. As you would expect a lot of DOW 30 stocks are close to fair value with the note the majority are now showing fully to overvalued, the strongest overvaluation we have seen in a while.

Website is StockCalc.com

DOW 30 Price v Value

Stockcalc – Free Trial

Have a free 30 day trial inside StockCalc www.StockCalc.com

The content presented here is in part from the website stockcalc.com and is provided on an as is or as available basis with all faults and may not be current in all cases. You should not rely on any of the information as authoritative or as a substitute for the exercise of your own skill and judgment in making an investment decision.

An Introduction to Stock Valuation

I have put together an overview of how stocks are valued using fundamental valuation methods.

How to value stocks

PDF version of this ebook: An Introduction to Stock Valuation

To run a valuation on a company try it here Get a valuation report

Background:

“Risk comes from not knowing what you are doing.” – Warren Buffet

Buying stocks without understanding their value is like buying a (car, set of golf clubs, vacation) without asking the (price, model, location) first. How do you know you are getting a good deal if you don’t know the value?

This e-book is an overview of valuation. Its purpose is to help you understand how to value stocks. Picking stocks, once you know this, is an easier process as it gives you a level of confidence that you are purchasing stocks that have a value you have determined based on the risks you understand.

The material can be a bit dry at times; we’ll try and keep it light.

Who should read this book?
This book is an introduction to valuation so there is some level of understanding that will be needed (and can easily be obtained). The book is of value:
– If you are investing but are not sure how the stocks you own are valued
– If you are aware of financial statements, may recognize Revenue and Net Income but not much else and want to expand that knowledge as it pertains to the Stock Market

There are many other parts to understanding what stocks to buy (or sell) and we will cover some of those in future editions, specifically analyzing a company’s financial statements and ratios to understand what risks we need to be aware of (does the company have too much debt compared to its industry, does the company face a liquidity crunch in its short term financing…) We do touch on ratios here and introduce some limited financial statement analysis.

About the Author:
Brian is the President of StockCalc (www.stockcalc.com) a fundamental valuation website for retail Investors and Investment Advisors. Brian is a Chartered Business Valuator (CBV), a Canadian valuation designation (www.cicbv.ca)

The full PDF version is available here – no email required. Feedback is very much appreciated.

PDF version of this ebook: An Introduction to Stock Valuation

Value a Stock in 4 Easy Steps

Determining what a stock is worth does not have to be a complicated process if you have the right tools to help you.  There are 4 steps to generating a valuation using the Stockcalc website. www.stockcalc.com

In the last blog we outlined the process involved
https://www.stockcalc.com/blog/BrianD/2015/12/21/how-to-calculate-what-a-stock-is-worth/

In this we will dig a little more into the 4 steps:  I am using the Stockcalc website to do these calculations and for full disclose am company President.

The 4 Steps (once you have selected a company)
1)  Determine what the cost is for the funds needed to run and grow the business. We call this the Weighted Average Cost of Capital or WACC.   Large, stable companies have a lower WACC than more speculative companies.
2) Forecast the company’s financials into the future based on assumptions you have or are able to get from Analysts that cover the company. We have a number of forecasting tools on the site starting with Analyst forecasts and Growth projections all the way to using a blank page and creating the forecast yourself.
3) Value the company using Valuation models such as a Discounted Cash Flow (DCF) where you include the financial forecasts, WACC and other calculations and assumptions such as Capital Expenditures needed and Debt levels.  The site has a full DCF framework for you to calculate with and auto-populates each cell to get you started.
4) Test your assumptions, see how sensitive the company is to the inputs.  Testing your assumptions is a critical part of valuation work. When you get a different valuation than you see a company is trading for on the stock exchange you need to ask why, and test. You may have uncovered an opportunity.

Here is a valuation I recently did for Lowes Companies Inc. using this 4 step process.
https://www.stockcalc.com/blog/BrianD/2015/12/21/lowes-companies-inc-low-fundamental-valuation-using-analyst-forecast-data/

And here is the process explained on Video for Alphabet (GOOG)
https://www.youtube.com/watch?v=V5194KeW_d0

Valuation is part art, part science.  The assumptions you make impact the company’s value.  For example, if you think the WACC is 8% instead of 9% the company will calculate to be more valuable because its cost to service its capital will be lower.

Here is a tool you can use for free to quickly test your assumptions :
www.stockcalc.com/dcf.aspx
Simply load a symbol or name into the Symbol text box and select the company from the dropdown.  Test the valuation by changing the growth rates, WACC, Free Cash Flows etc.

Next Steps:
If you are not sure where to begin you can select a company you are familiar with (GOOG, AAPL) and work though the steps above on the Stockcalc site.   Each of these steps are found on the Research Page which you can access either by clicking the Research Button   or selecting Research from the dropdown menu next to the Stockcalc logo (both are on the Dashboard)

About Stockcalc:
If you would like to explore the Stockcalc website and quickly run valuations like  simply create an account at www.stockcalc.com  (Start with a 14 day free trial)  Use the walk-throughs (click the walking man icon), videos (video icon on each page) or the help menu to help navigate the site. The site has a number of tools for data query, backtesting,  forecasting and valuation.  We have a no restrictions  Stockcalc 14 Day Free Trial available as well.
If you would like the above valuation to test simply send us a note from Stockcalc’s “Contact Us” on the dashboard.

 

 

How sensitive are companies to rising interest rates?

The talk of will they or won’t they raise rates has been ongoing for more than 2 years now.  We have to expect either rates get raised at some point, we stumble along on the current path or we end up in deflationary times.  I want to explore the more optimistic of these scenarios and its impact on stock valuations: the economy improves and rates start to rise.

To look at this in detail I am using www.stockcalc.com  (Disclaimer: I am the President of Patchell Brook Equity Analytics and we created Stockcalc.com precisely for this type of analysis and valuation)

I will examine Caterpillar as it has with a 59/41 Equity to Debt in the Capital Structure

Using Stockcalc I first create and save a WACC for CAT:  (Research Menu, Valuation, WACC).  Stockcalc does a first pass on WACC using default values which results in a WACC of 8.93.

Cost of Equity (Ke) 13.63 (Using CAPM), After Tax cost of Debt : 2.17, 59% Equity, 41% Debt

 

WACC for CAT

Next I load the Quick DCF tool  so I can run the analysis (Research, Forecast, Analyst) (Note – the quick DCF is a testing and teaching tool, the other tools are more detailed in their analysis)

It is available for free at www.stockcalc.com/dcf.aspx or inside the Stockcalc.Com website

Quick DCF for CAT

Value Per Share ($) Calculations for Caterpillar (CAT:NYS) using the Quick DCF tool on Stockcalc.com

Each change in WACC (8.93 – 9.34 – 9.75) implies a 1 % change in Cost of Debt.  Growth rates are next 3 years and thereafter (5% and 3% in the first row for ex)

WACC 8.93 9.34 9.75
Growth rates 5,3% 70.02 61.28 53.60
Growth rates 6,4% 97.40 84.92 74.21
Growth rates 4,2% 50.55 44.08 38.30

(Current Price at the time of valuation (October 2015) was $69.34)

A 1% change in the Cost of Debt results in a 12.5% change in value per share.  A 1% change in Growth rates results in a 28% change in Value per Share.  A 1% change in WACC results in a similar 28% change in Value per Share.

Message: if interest rates are rising, growth prospects need to be rising at half the rate to keep this share price holding steady.

 

If you want to explore the Stockcalc software simply create an account at www.stockcalc.com and have a look around.  Use the walk-throughs (click the walking man icon), videos (video icon on each page) and  help menu to understand and navigate the site.  The site contains a number of tools including screens, queries, backtests, forecasts and a variety of valuation methods.