WhitmanTrading

Weighted Average: Ask What the Weight Was

A weighted average multiplies each value by an importance factor before averaging, so larger contributions count for more. It underlies average entry price, volume-weighted average price, diluted share counts and the cost of capital, and choosing the weight is the only real decision in it.

How it works

A labelled statement diagram showing 100 shares at 10 and 900 shares at 20 producing a weighted average of 19. The headline reads: An average where some numbers count for more than others.
An average where some numbers count for more than others. Illustrative figures - not a real company.

A weighted average multiplies before it averages. One hundred shares at 10 and nine hundred at 20 give a weighted average of 19, because the second purchase is nine times the size of the first.

A labelled statement diagram comparing a simple average of 15 with a weighted average of 19. The headline reads: The simple average of the same two prices is 15.
The simple average of the same two prices is 15. Illustrative figures - not a real company.

The simple average of the same two prices is 15. Four units of difference, from identical inputs, purely because one method counted the sizes and the other did not.

A labelled statement diagram comparing a small position of 100 with a large one of 900. The headline reads: The weight is the point, and choosing it is the decision.
The weight is the point, and choosing it is the decision. Illustrative figures - not a real company.

Choosing the weight is the only real decision. Size, time, volume and market capitalisation are all defensible weights for different questions, and they produce different answers to the same one.

Where traders meet it

A labelled statement diagram showing three purchases at 100, 104 and 110 giving an average entry of 105. The headline reads: It is how a scaled entry's real cost is calculated.
It is how a scaled entry's real cost is calculated. Illustrative figures - not a real company.

A scaled entry has a weighted average cost. Three equal thirds at 100, 104 and 110 give 105 — and if the thirds were not equal, the answer moves toward whichever tranche was largest.

A labelled statement diagram showing a purchase at 100, another at 80, and a new average of 90. The headline reads: And how averaging down lowers the price and raises the risk.
And how averaging down lowers the price and raises the risk. Illustrative figures - not a real company.

Averaging down improves the number and worsens the position. Buying again at 80 after buying at 100 lowers the average to 90, which looks like progress on a screen.

A labelled statement diagram showing an original position and an addition doubling total exposure. The headline reads: The position doubled while the average improved by ten.
The position doubled while the average improved by ten. Illustrative figures - not a real company.

Meanwhile the exposure doubled. The average price fell ten per cent and the amount at risk rose a hundred per cent, and only one of those two numbers is displayed prominently. That asymmetry is why averaging down feels better than it is.

A labelled statement diagram comparing a session's simple average price with its volume-weighted average. The headline reads: The volume-weighted average price is the same idea on a chart.
The volume-weighted average price is the same idea on a chart. Illustrative figures - not a real company.

VWAP is this calculation applied to a session. Every trade weighted by its size, giving the average price at which the day’s volume actually changed hands.

In practice: where else it appears

A labelled statement diagram showing 20 shares for three months and 25 for nine, giving a weighted average of 24. The headline reads: And earnings per share uses a weighted average share count.
And earnings per share uses a weighted average share count. Illustrative figures - not a real company.

Earnings per share uses a time-weighted share count. Shares issued in month four count for three quarters of the year, which is why the denominator rarely matches the year-end figure.

A labelled statement diagram showing earliest and latest stock costs producing a weighted average cost. The headline reads: Inventory can be valued the same way, and often is.
Inventory can be valued the same way, and often is. Illustrative figures - not a real company.

Inventory is frequently valued this way. Weighted average cost is one of the permitted methods, and it produces a different cost of goods sold — and therefore a different profit — from first-in-first-out.

A labelled statement diagram showing a cost of debt of five per cent and a cost of equity of ten producing a weighted average of eight. The headline reads: And the cost of capital is a weighted average of debt and equity.
And the cost of capital is a weighted average of debt and equity. Illustrative figures - not a real company.

The cost of capital is a weighted average too. Debt and equity each carry a rate, weighted by how much of each the company uses — which is the discount rate a valuation model needs.

A labelled statement diagram showing the lowest and highest prices paid with an average between them. The headline reads: A weighted average hides the spread it was built from.
A weighted average hides the spread it was built from. Illustrative figures - not a real company.

Every average destroys information. An average entry of 90 could come from two purchases at 80 and 110 or twenty purchases between 88 and 92, and those are completely different positions.

A labelled statement diagram comparing a size-weighted average of 19 with an equally weighted one of 15. The headline reads: Always ask what the weight was before trusting the average.
Always ask what the weight was before trusting the average. Illustrative figures - not a real company.

So the question to ask of any average is what was weighted. An index weighted by market capitalisation and one weighted equally describe the same companies and behave completely differently.

One case deserves separating because the two averages answer genuinely different questions: an index. A market-capitalisation-weighted index tells you what the market as a whole did, because it reflects how much money is actually invested in each company. An equally weighted version tells you what the average company did, which is a different and equally legitimate question.

The two diverge most when a few very large constituents move. A handful of enormous companies rising while most fall produces a rising capitalisation-weighted index and a falling equal-weighted one — both correct, describing the same day. When the two disagree sharply, the disagreement is the story, and it is a reading nobody gets from either number alone.

The same logic applies to a portfolio. Return weighted by position size says what your money did; return weighted equally says what your selections did. A large gap between them says your sizing, rather than your picking, is doing most of the work — in either direction, and that is worth knowing before changing how you choose what to buy.

What a weighted average is not

It is not a simple average. The two disagree whenever sizes differ.

It is not neutral. The weight is a choice that changes the answer.

It is not a complete description. The spread is gone.

And it is not always the right average. Sometimes equal weighting is.

When it fails as a summary

A weighted average is dominated by its largest component. Nine hundred of one thing and one hundred of another produces a figure that is essentially a description of the nine hundred, and the smaller contribution is invisible.

A second failure is a moving average entry price used as a performance measure. It improves whenever you add at a lower price, so it rewards exactly the behaviour that increases risk.

A third is comparing weighted figures computed on different weights. Two index returns weighted differently are not comparable, however similar the constituents.

A fourth is a time-weighted share count read as a share count. The year-end figure is what you own a fraction of; the weighted one is only for the earnings calculation.

And a fifth is trusting an average without the range. The spread behind it is what tells you whether the average describes anything.

The original data

Of the 31,760 trading and investing videos in this site’s corpus, “valuation” returns 9 at a median of 17,868 views, “financial statements” 3 at a median of 1,065,893, and “cash flow” 17 at a median of 67,134 across 14 channels. The counts are in research/corpus-coverage.json, produced by site/measure_corpus.py.

The figures in the diagrams on this page are illustrative. The 15 against 19 from the same two prices is the whole idea in one comparison, and the 100-to-900 split is deliberately extreme to make the effect visible — real weightings are usually less lopsided and work in exactly the same way.

The habit worth keeping is to record the range alongside the average, in a position and in a set of accounts. Highest paid, lowest paid, and the weighted average between them takes one extra line and preserves the information the average removes — which is the difference between knowing your cost and knowing your position.

VWAP is the intraday version used as a reference level. Position sizing is where the weight in a scaled entry gets decided. And financial statements is where the accounting versions appear.

What I actually do

The moment this stopped being arithmetic and started being useful was when I noticed my average entry price was flattering me. It had improved because I had added to a losing position, which is the one circumstance where a better average price means a worse situation.

— Michael Whitman

This page is educational, not financial advice. Test every idea on your own charts before risking money.