Full and sampled replication
In a full educational replication, the fund holds the constituents at the required weights. With sampling, it holds only part of the set or another representative structure in an attempt to approximate the characteristics. Sampling can produce incomplete matching, but the method name does not calculate the future gap. Transaction costs, constituent availability, weight-change dates and cash balances matter. Derivatives, when allowed by the documents, add obligations that must be understood rather than treated as direct ownership.
Suppose a fictional index has two components weighted 60% and 40%, with returns of +10% and −5% for a period. With the initial weights, the index return is 4%. A teaching portfolio holding only the first component earns 10% before costs. That is not evidence that sampling is superior: a different move in the second component could reverse the sign.
Passive naming does not remove costs
If a full portfolio produces 4% before cost and hypothetical period costs are 0.5% of the starting base, the simplified net result is 3.5%; fund minus index is −0.5 percentage point. The linear subtraction is an explicit teaching convention, not the fee-accrual method of a particular product. Fund owners bear expenses regardless of result, so read them in the documents instead of inferring them from the word “index”.
Cash balances, distribution dates, rebalancing and transaction costs can also change results. Distinguish a fund’s NAV movement from the result of an investor who bought at a premium or sold at a discount to NAV. A comparison of the index with one transaction price can mix tracking with entry and exit costs if the time bases differ. Decide which object is being compared first.
Two measures of deviation
Tracking difference usually means the fund result minus the index result over an agreed period, with the sign stated. Tracking error is the variability of the deviation series; its estimate and periodicity must be specified. One annual number cannot establish the stability of monthly deviations. In a demonstration, take four monthly deviations: +0.1, −0.1, +0.2 and −0.2 percentage points. The mean is zero, while the population standard deviation is approximately 0.1581 percentage point. A zero average therefore does not mean exact matching each period.
This is arithmetic for the stated series, not a universal methodology: a sample estimate or annualisation would give different values. Real comparisons need the same dates, currencies, distributions, costs and calculation rules. The guide does not rank products, recommend investing from past deviation or treat passive packaging as a capital guarantee.
Control the chosen benchmark
If the index includes reinvested distributions but the fund series excludes them, the observed lag cannot all be assigned to management. Likewise, a currency move must not enter only one side. Align the economic content and dates first, then calculate the difference. A visually attractive chart does not prove tracking quality.
For both deviation measures, state the sign, frequency and variability estimator. A zero average can coexist with noticeable swings, while a small persistent lag can have low variability. Neither measure is a complete assessment of fund risk. Do not rank products by one number without the asset mix, costs and a comparable benchmark.
The capitalisation filter comes before weighting
A size index first defines the eligible company set: market and security categories, trading and information criteria, price and share-count dates. It may then rank companies by full capitalisation, select a size segment and only afterwards set weights, for example with a free-float adjustment. FTSE Russell methodology distinguishes full-capitalisation ranking from the investable fraction and describes reviews and buffers around boundaries. This is a methodology example, not universal small/mid/large thresholds.
In an author-created model, five eligible companies have capitalisations of 100, 60, 30, 8 and 2 billion units. A “two largest” filter selects the first two, with full-capitalisation weights of 62.5% and 37.5%. If free float is 40% and 100%, investable capitalisations of 40 and 60 give weights of 40% and 60%. Selection and weighting are different operations.
A fictional filter includes companies from 10 to 50 billion on the review date. A company valued at 9 is excluded; after rising to 11 it passes the ordinary threshold at the next review. If an existing constituent exits below 8 while a new entrant must reach 12, the same value of 11 receives different treatment based on prior status. That is an explicit buffer example, not a current-index rule.
A monetary threshold and a ranked-place filter differ: broad market growth can move many companies across a fixed boundary while the number of ranked places stays constant. Review dates, effective dates, corporate events and buffer rules belong in the methodology. A fund then implements the set with costs and possible delay. Capitalisation growth does not imply immediate inclusion or a mechanical price increase.