Estimating the regression line and interpreting coefficients

Estimating the regression line and interpreting coefficients

Estimating the regression line and interpreting coefficients

On this page

Learning Outcome Statement

Describe, interpret, and explain simple linear regression, including coefficient estimation using the least squares criterion.

On the exam: The question gives paired periodic observations in a table and asks you to estimate the least-squares intercept and slope, identify the fitted line, or interpret a coefficient in a fill-in, multiple-choice, or ordered-response format.

Simple linear regression uses a single independent variable to explain a dependent variable with a straight line. In pair $i$, $x_i$ is the independent observation and $y_i$ is the dependent observation. The line turns $x_i$ into a fitted value, $\hat{y}_i$, and the residual, $e_i$, is actual $y_i$ minus that fitted value. Squaring each residual stops positive and negative misses from canceling and gives larger misses more weight.

SymbolDefinitionUnit
$x_i$independent observation$x$ units
$y_i$dependent observation$y$ units
$\hat{y}_i$fitted value from line$y$ units
$e_i$actual minus fitted$y$ units
$\bar{x}$sample mean of $x$$x$ units
$\bar{y}$sample mean of $y$$y$ units
$n$paired observationsobservations
$\hat{\beta}_0$fitted intercept$y$ units
$\hat{\beta}_1$fitted slope$y$ units per $x$ unit

The sample means provide the center of the observations, so centering separates the line’s tilt from its location. The coefficient units follow from the ratio in the slope and the direct level in the intercept.

Least squares chooses the intercept and slope that minimize the sum of squared residuals. With $\mathrm{SSE}$ naming that sum, the fitted line and coefficient estimates are:

$$ \begin{aligned} \hat{y}_i &= \hat{\beta}0+\hat{\beta}1x_i \ \mathrm{SSE} &= \sum{i=1}^{n}e_i^2=\sum{i=1}^{n}(y_i-\hat{y}i)^2 \ \hat{\beta}1 &= \frac{\sum{i=1}^{n}(x_i-\bar{x})(y_i-\bar{y})}{\sum{i=1}^{n}(x_i-\bar{x})^2} \ \hat{\beta}_0 &= \bar{y}-\hat{\beta}_1\bar{x} \end{aligned} $$

The numerator of the slope rewards pairs whose deviations from their means move in the same direction. Opposite deviations pull the slope down. The denominator records how widely $x$ moves, so more centered spread in $x$ makes the same co-movement produce a smaller slope. The intercept places the line so it passes through $(\bar{x},\bar{y})$, and it is fitted $y$ when $x$ equals zero.

Use this sequence when the table gives paired observations. Before calculating, keep the pairs intact: each $x_i$ belongs with its observed $y_i$. Find both sample means, then measure each observation relative to its own mean. Those centered deviations reveal whether the variables move together and how much independent-variable spread supports the fitted tilt. Once the slope is known, the mean point fixes the intercept and gives you a check on the line.

Loading diagram…

The geometry is simple: observed points sit away from a fitted line, and residual arrows measure the gaps. Each arrow is a vertical difference at a particular observed $x$, measured from actual $y$ to fitted $\hat{y}$. A point above the line has a positive residual; one below has a negative residual. Least squares squares these signed gaps before adding them, so neither direction cancels the other and a larger miss contributes more.

Worked example

You are handed paired monthly observations from a market excess-return series and a stock excess-return series. $x$ is the market excess return and $y$ is the stock excess return, both measured in percentage points per month. Treat the observations as sample data, keep the monthly frequency, and do not annualize the inputs. You need the least-squares intercept, slope, and fitted line.

GivenValue
$n$$5$ paired observations
$x$$-2.0000, -1.0000, 1.0000, 2.0000, 3.0000$ percentage points per month
$y$$-1.0000, 0.0000, 2.0000, 1.0000, 4.0000$ percentage points per month
FrequencyMonthly
Data treatmentSample data, no annualizing

The entries in the $x$ and $y$ rows pair by position. The observed series is $y$, and the fitted series is $\hat{y}$ at the same $x$ values. Keep the units and frequency attached to both series throughout the calculation. The diagram gives a quick visual comparison: the first line traces the observed returns, while the second shows the candidate fitted values across the same ordered market returns. Vertical separation at any position is the residual for that pair. This makes the sign and size of each miss easier to inspect before working through the coefficient arithmetic.

Loading diagram…

Read the plotted series at each shared $x$ value rather than comparing points across different positions. The observed line rises and falls with the sample, while the fitted line follows the constant slope implied by the estimated coefficients. For example, the fitted value at the leftmost market return is below the observed return, and at the final return it is above. The following rows make those pairwise comparisons exact and let you check the fitted line numerically.

$x$Observed $y$Fitted $\hat{y}$
$-2.0000$$-1.0000$$-4.0000$
$-1.0000$$0.0000$$-2.0000$
$1.0000$$2.0000$$2.0000$
$2.0000$$1.0000$$4.0000$
$3.0000$$4.0000$$6.0000$
  1. Average each column to anchor the line at the sample center. The totals are $3.0000$ and $6.0000$, so $\bar{x}=\frac{3.0000}{5}=0.6000$ and $\bar{y}=\frac{6.0000}{5}=1.2000$.

  2. Subtract each mean before multiplying, because least squares uses deviations rather than raw levels. The centered cross-product sum is $14.4000$, and the centered $x$-deviation square sum is $7.2000$.

  3. Divide centered co-movement by centered $x$ spread to get the slope. $\hat{\beta}_1=\frac{14.4000}{7.2000}=2.0000$.

  4. Use the mean point to anchor the line and obtain the intercept. $\hat{\beta}_0=1.2000-2.0000\times0.6000=0.0000$.

  5. Insert both coefficients to identify the fitted line. $\hat{y}=0.0000+2.0000x$.

The fitted line is $\hat{y}=0.0000+2.0000x$. For the decision, read the slope as a fitted change in $y$ for a one-unit change in $x$, and read the intercept as fitted $y$ when $x$ equals zero.

Changed condition

The same monthly setup changes only the last observed $y$. Before looking at the new coefficients, predict a higher slope. The changed observation sits at the largest positive $x$, so raising its $y$ strengthens centered co-movement while leaving the centered $x$ spread unchanged.

QuantityBeforeAfter
Last $y$$4.0000$$5.0000$
$\hat{\beta}_1$$2.0000$$2.3333$
$\hat{\beta}_0$$0.0000$$-0.0000$

The prediction is correct. The slope rises, while the intercept stays at the sample-center anchor because the fitted line passes through the sample means.

Before moving on, redo the centered numerator and denominator yourself.

Common error

A candidate may calculate the slope from uncentered products or divide by the number of observations. It feels efficient because that resembles a covariance or variance shortcut built from the table’s raw entries. Both variables have nonzero means, so the deviations must be centered around their sample means before the products are summed, and the centered sums are not divided by the number of observations.

Calculator

Enter each paired observation in the data editor before asking the calculator for the slope. The intercept still comes from the mean-anchor formula.

KeysDisplay
2ND CLR WORK0.0000
2ND DATA0.0000
-2.0000 ENTER DOWN -1.0000 ENTER-1.0000
-1.0000 ENTER DOWN 0.0000 ENTER0.0000
1.0000 ENTER DOWN 2.0000 ENTER2.0000
2.0000 ENTER DOWN 1.0000 ENTER1.0000
3.0000 ENTER DOWN 4.0000 ENTER4.0000
2ND STAT LIN0.0000
CPT b2.0000

Check yourself

Practice

Which interpretation matches the fitted slope of 2.0000 in the worked example?

Put the coefficient-estimation steps in order.

  • Pair each x with its y
  • Compute sample means
  • Center deviations and form sums
  • Divide centered sums for slope
  • Use means for intercept
  • Interpret fitted coefficients

Revision snapshot

Least squares chooses the intercept and slope that minimize the sum of squared residuals. Use centered observations:

$$ \hat{\beta}1=\frac{\sum{i=1}^{n}(x_i-\bar{x})(y_i-\bar{y})}{\sum_{i=1}^{n}(x_i-\bar{x})^2} \qquad \hat{\beta}_0=\bar{y}-\hat{\beta}_1\bar{x} $$

The slope is the fitted change in $y$ for a one-unit change in $x$. The intercept is fitted $y$ when $x$ equals zero. Raw products or division by the number of observations are the trap when both means are nonzero. On the exam, keep the paired periodic inputs in their stated frequency, estimate the line, and interpret the coefficient in its units.

Unlock the full lesson and continue reading

Signup to continue reading this lesson and unlimited access questions, flashcards, AI notes, and more

Scan to download app

Scan to download
UNLOCK FREE ACCESS

Practice Questions: Estimating the regression line and interpreting coefficients

  1. 1. A manager reports performance after transaction costs but before deducting management fees. The reported measure is best described as:

    • A. Real return
    • B. Gross return
    • C. Net return
    Show answer

    Answer: B. Gross return

    > Key point: Gross return includes transaction costs but is reported before management fees. Why Gross return is correct - The reported performance is after transaction costs and before management fees, which matches gross return. How they compare | Measure | Treatment | |---|---| | Gross return | After transaction costs; before management fees | | Net return | After management fees | | Real return | Adjusted for inflation | Why the other options are wrong - Net return: Management fees have not been deducted. - Real return: No inflation adjustment is described.

  2. 2. An analyst calculates a stock's holding-period return using only its price change, although the investor received a cash dividend during the period. The reported return most likely:

    • A. Understates the investor's holding-period return
    • B. Correctly measures the investor's holding-period return
    • C. Overstates the investor's holding-period return
    Show answer

    Answer: A. Understates the investor's holding-period return

    > Key point: Holding-period return includes both the stock’s price change and any cash dividend received. Why understates the investor’s holding-period return is correct - A price-only calculation omits the dividend from the investor’s total return. - The full return includes the cash dividend as well as the price change. Why the other options are wrong - Correctly measures the investor’s holding-period return: it measures price return only, not total holding-period return. - Overstates the investor’s holding-period return: omitting the dividend makes the reported return lower, not higher.

  3. 3. An investor will receive five annual payments of USD 20,000, with the first payment today. At an effective annual discount rate of 6%, the calculated present value is closest to:

    • A. USD 84,247
    • B. USD 89,302
    • C. USD 94,660
    Show answer

    Answer: B. USD 89,302

    > Key point: A payment received today is discounted for zero years, so include it at full value. Why USD 89,302 is correct - The five payments occur at times 0 through 4. - Present value = USD 20,000 + USD 20,000 × [1 − (1.06)⁻⁴] / 0.06 = USD 89,302. Working | Step | Working | Result | |---|---|---:| | Payments at times 0–4 | USD 20,000 × [1 + (1 − 1.06⁻⁴) / 0.06] | USD 89,302 | | Ordinary annuity | USD 20,000 × [1 − 1.06⁻⁵] / 0.06 | USD 84,247 | | One year too early | USD 89,302 × 1.06 | USD 94,660 | Answer: USD 89,302 Why the other options are wrong - USD 84,247: Treats the first payment as arriving in one year, rather than today. - USD 94,660: Compounds the correct value by one extra year, placing the payments one year too early.

  4. 4. A return sample has a mean of 3.6%, median of 2.4%, and mode of 1.5%. Its distribution is most likely:

    • A. Negatively skewed
    • B. Approximately symmetric
    • C. Positively skewed
    Show answer

    Answer: C. Positively skewed

    > Key point: Mean > median > mode is the characteristic ordering of a positively skewed distribution. Why Positively skewed is correct - The sample has mean 3.6% > median 2.4% > mode 1.5%. - A longer right tail pulls the mean above the median and mode. How they compare | Distribution | Typical relationship | |---|---| | Positively skewed | Mean > median > mode | | Negatively skewed | Mean < median < mode | | Approximately symmetric | Mean, median, and mode are close | Why the other options are wrong - Negatively skewed: Its typical ordering is mean < median < mode, the reverse of the sample. - Approximately symmetric: The three reported measures are not close together; they show a clear ordering.

  5. 5. A bank deposit pays a stated annual rate of 6.00%, compounded monthly. The calculated value of its effective annual rate is closest to:

    • A. 6.00%
    • B. 6.17%
    • C. 6.14%
    Show answer

    Answer: B. 6.17%

    > Key point: Convert the stated rate to a monthly rate, compound for 12 months, then subtract 1. Why 6.17% is correct - Monthly rate = 0.06 ÷ 12 = 0.005. - Effective annual rate = (1.005)¹² − 1 = 0.0616778, or 6.17% rounded. Working | Step | Working | Result | |---|---|---| | Monthly rate | 0.06 ÷ 12 | 0.005 | | Compound for one year | (1 + 0.005)¹² − 1 | 0.0616778 | | Convert to percentage | 0.0616778 × 100 | 6.17% | Answer: 6.17% Why the other options are wrong - 6.00%: This is the stated annual rate; adding the monthly rates without compounding gives 6.00%. - 6.14%: This results from quarterly compounding, (1 + 0.06 ÷ 4)⁴ − 1, but the deposit compounds monthly.

  6. 6. A risk analyst models continuously compounded returns as normally distributed. The resulting distribution of the asset's future price is best described as:

    • A. Symmetric with no lower bound
    • B. Right-skewed with a lower bound of zero
    • C. Left-skewed with an upper bound of zero
    Show answer

    Answer: B. Right-skewed with a lower bound of zero

    > Key point: Normally distributed continuously compounded returns imply a lognormal future price, which is right-skewed and bounded below by zero. Why Right-skewed with a lower bound of zero is correct - Future price is the exponential of the continuously compounded return, so it is positive. - The resulting lognormal distribution has a long right tail. How they compare | Description | What it describes | Why it fits or fails | |---|---|---| | Right-skewed; lower bound zero | Future price | Lognormal: positive and right-skewed | | Symmetric; no lower bound | Normal return | Describes returns, not price | | Left-skewed; upper bound zero | Neither | Reverses the skew and the bound | Why the other options are wrong - Symmetric with no lower bound: describes the normally distributed return; applying it to price would allow negative values. - Left-skewed with an upper bound of zero: reverses both features; the price is right-skewed and has a lower bound of zero.

  7. 7. An analyst trains an algorithm on loans labeled default or non-default to classify new loans. This approach is best described as:

    • A. Supervised learning
    • B. Unsupervised learning
    • C. Reinforcement learning
    Show answer

    Answer: A. Supervised learning

    > Key point: Learning from examples labeled with known outcomes is supervised learning. Why Supervised learning is correct - The training loans are labeled default or non-default, so the algorithm learns from known outcomes. - It uses those examples to classify new loans. How they compare | Learning approach | Training signal | Why it fits or not | |---|---|---| | Supervised learning | Labeled examples | Learns to predict the loan categories | | Unsupervised learning | Unlabeled data | Finds patterns without known outcome labels | | Reinforcement learning | Rewards or penalties | Learns actions from feedback, not labeled loan examples | Why the other options are wrong - Unsupervised learning: the training loans have known default outcomes, rather than unlabeled data. - Reinforcement learning: the algorithm is not choosing actions based on rewards or penalties.

  8. 8. An analyst compares the arithmetic mean return and the geometric mean return calculated from the same series of historical periodic returns. The geometric mean return is most likely:

    • A. Equal to the arithmetic mean return only when returns are identical in every period
    • B. Greater than the arithmetic mean return whenever returns vary from period to period
    • C. Less than the arithmetic mean return whenever returns vary from period to period
    Show answer

    Answer: C. Less than the arithmetic mean return whenever returns vary from period to period

    > Key point: When periodic returns vary, the geometric mean return is less than the arithmetic mean return. Why less than the arithmetic mean return whenever returns vary from period to period is correct - The arithmetic–geometric mean inequality applied to periodic growth factors makes the geometric mean no greater than the arithmetic mean. - If returns vary, the inequality is strict, so the geometric mean is lower. Why the other options are wrong - Equal to the arithmetic mean return only when returns are identical in every period: This states the equality condition, not the relationship when returns vary. - Greater than the arithmetic mean return whenever returns vary from period to period: Variation makes the geometric mean lower, not higher.

Flashcards: Estimating the regression line and interpreting coefficients

  1. 1. What is the purpose of simple linear regression?

    Show answer

    To explain the variation in a dependent variable using a single independent variable.

  2. 2. In simple linear regression, what does the slope (\(b1\)) represent?

    Show answer

    The expected change in the dependent variable for a one-unit change in the independent variable.

  3. 3. What is the least squares (OLS) criterion in linear regression?

    Show answer

    The regression line is the one that minimizes the sum of squared errors (SSE).

  4. 4. How is \(R^2\) calculated specifically in a simple linear regression?

    Show answer

    By squaring the correlation coefficient: \[R^2 = r^2\]

  5. 5. What formula determines the prediction interval for a regression forecast?

    Show answer

    \[\hat{Y} \pm (tc \times sf)\] where \(sf\) is the standard error of the forecast.

  6. 6. In a Lin-Log regression model, how is the slope coefficient interpreted?

    Show answer

    The absolute change in Y for a relative (percentage) change in X.

  7. 7. What does the Coefficient of Determination (\(R^2\)) represent?

    Show answer

    The percentage of total variation in the dependent variable explained by the independent variable.

  8. 8. What is the mathematical relationship between SST, SSR, and SSE?

    Show answer

    \[SST = SSR + SSE\]

  9. 9. What does the Total Sum of Squares (SST) measure in regression?

    Show answer

    The total variation in the dependent variable around its mean.

Everything you need for CFA prep

Get full Oncourse access with lessons, practice questions, flashcards and AI study tools.

GET STARTED FOR FREE
75% OFFLimited time offer
GET 75% OFF