Draft:Bar Construction (Financial Data)

Not to be confused with Bar (technical analysis).

Bar construction is a group of data-aggregation techniques used in quantitative finance to transform high-frequency trading data into regularly structured observations known as bars. Bars summarize market activity over a defined quantity or interval and are used in empirical finance, econometrics and market microstructure research for tasks such as volatility estimation, price-impact analysis and algorithmic trading system design.

High-frequency data, such as records of individual trades or quotes, are typically observed at irregular time intervals. Bar-construction methods provide rules for converting these irregular sequences into series of observations sampled according to clock time, transaction count, traded volume or traded notional value. Different choices of sampling rule have been shown to result in different empirical properties of returns, including their distribution, autocorrelation and volatility dynamics.

Background

Research on high-frequency financial data often treats prices as evolving in "business time" or "transaction time" rather than physical time. Clark (1973) proposed modelling speculative prices as a subordinated stochastic process, where returns over calendar time are driven by an underlying activity process such as trading volume. Under this view, observing returns at constant calendar intervals may mix periods of high and low market activity, which can lead to volatility clustering and heavy-tailed distributions.

Subsequent work has examined alternative clocks based on transactions or volume. Mandelbrot and Taylor (1967) studied price changes measured over fixed numbers of transactions and compared them with changes measured over fixed time intervals. Ané and Geman (2000) analysed transaction-time models and documented that, for several assets, returns sampled on a transaction clock are closer to normal than returns sampled on a calendar clock.

These studies have motivated procedures that aggregate high-frequency data according to measures of trading activity. In practice, bar construction is used both in academic work and in industry systems to obtain time series with more homogeneous spacing in the chosen sampling variable.

Data representation

Let the high-frequency trade record for a single financial instrument be represented as a sequence

$\{(t_i,p_i,v_i){\}}_{i=1}^{\mathrm{\infty }}$,

where $t_i$ denotes the time stamp of trade $i$, $p_i$ is the transaction price and $v_i$ is the traded quantity (for example, number of shares or contracts). The objective of bar construction is to partition this sequence into non-overlapping subsets, each corresponding to a bar, and to compute summary statistics for each subset.

Common bar statistics include:

• open price (price of the first trade in the bar);
• high and low prices (maximum and minimum trade prices in the bar);
• close price (price of the last trade in the bar);
• total traded volume and notional value;
• derived quantities such as volume-weighted average price (VWAP) and realized variance.

The sampling rule determines how many trades belong to each bar and when the bar closes.

Standard bar types

Several bar types are widely described in the literature and in practitioner texts. They differ in the variable used as the sampling clock.

Time bars

Time bars aggregate all trades that occur within fixed chronological intervals, such as one second, one minute or five minutes. If $\Delta$ denotes the length of the interval and $k$ is an integer, the $k$-th time bar consists of all trades with time stamps in $[k\Delta ,(k+1)\Delta )$. Time-bar data are common in commercial data feeds and in empirical studies using intraday prices at regular calendar intervals.

The main advantage of time bars is simplicity: they are easy to construct and align naturally with trading sessions, opening and closing times and other calendar-based events. However, because trading intensity varies over the trading day and across days, the number of trades and the amount of volume contained in each time bar can vary substantially. Several authors have reported that returns sampled at fixed time intervals exhibit volatility clustering and non-Gaussian features that are partly attributable to this variation in activity.

Tick bars

Tick bars (or transaction bars) consist of a fixed number of trades. For a threshold $N_{\text{ticks}}$, the bar closes after $N_{\text{ticks}}$ transactions have been observed. Formally, if $i_s$ is the index of the last trade of the previous bar, the index $i_e$ of the current bar is

$i_e=i_s+N_{\text{ticks}}.$

Sampling in transaction time aims to give each bar an equal number of price-discovery events. Empirical work has examined how the distribution of returns changes when data are viewed in transaction time instead of calendar time. Tick bars are used in some market microstructure studies and in trading systems that model price changes per trade.

In practice, tick bars may be influenced by market structure features. For example, exchange opening and closing auctions can be reported as single trades representing many underlying orders, and algorithmic execution strategies may fragment a single economic order into multiple small trades.

Volume bars

Volume bars aggregate trades until cumulative traded quantity reaches a specified threshold $V^{*}$. If the previous bar ends at index $i_s$, the current bar includes the smallest index $i_e$ such that

$\sum _{i=i_s+1}^{i_e}{v}_i\ge V^{*}.$

Volume-based sampling is closely related to subordinated-time models, in which trading volume is treated as a measure of business activity. In these models, volume plays the role of an operational time that drives the evolution of prices. Volume bars therefore attempt to allocate roughly similar amounts of traded quantity to each observation, which can be useful in studies of price impact and order-flow dynamics.

Volume bars appear in both academic and practitioner research as a way to regularise data when the size of trades is considered important, for example in equity and futures markets where order sizes vary significantly.

Dollar (value) bars

Dollar bars (or value bars) close when cumulative traded notional value reaches a predetermined threshold $D^{*}$. Notional value is defined as the product of trade price and quantity, $p_i{v}_i$. The bar ending index $i_e$ is the smallest index such that

$\sum _{i=i_s+1}^{i_e}{p}_i{v}_i\ge D^{*}.$

Using notional value as the sampling clock aims to allocate roughly equal economic value, rather than equal volume or equal trade count, to each bar. This can be useful when comparing periods with different price levels or when corporate actions such as stock splits change the relationship between share count and market value. Practitioner sources report that, for some instruments, dollar bars produce a relatively stable number of bars per trading day over long sample periods.

Comparison of bar types

The main bar types can be summarised as follows:

Method	Sampling criterion	Typical uses	Considerations
Time bar	Fixed chronological interval	Widely used in intraday charts and volatility estimation	Content per bar depends on trading intensity; may display strong intraday seasonality.
Tick bar	Fixed number of trades	Studies of transaction-time dynamics and order flow	Sensitive to order fragmentation and special auction mechanisms.
Volume bar	Fixed traded quantity	Analyses that focus on trade size and liquidity	Requires reliable volume data; irregular spacing in illiquid instruments.
Dollar bar	Fixed traded notional value	Applications comparing periods with different price levels or volatility regimes	Slightly more complex to compute; depends on both price and volume.

No single bar type is universally preferred; the choice depends on the research question, the characteristics of the instrument and the availability of high-frequency data.

Information-based bar methods

In addition to fixed-threshold bars, some practitioner literature describes adaptive or information-based bar-construction methods. These aim to define bar boundaries using measures of order-flow imbalance or run length rather than simple counts or totals.

One class of methods, often referred to as imbalance bars, accumulates signed trade volume (for example, buyer-initiated minus seller-initiated volume) and closes the bar when the cumulative imbalance deviates from its estimated expected value by a specified amount. Another class, sometimes called run bars, closes the bar when the number of consecutive buyer-initiated or seller-initiated trades exceeds a threshold based on historical run lengths.

These approaches seek to construct bars that capture similar amounts of directional order-flow information. They have been proposed for use in microstructure studies and high-frequency trading systems; however, they are less documented in peer-reviewed academic literature than standard time, tick, volume and dollar bars.

Implementation considerations

Practical implementation of bar construction involves several technical choices:

• Data source and quality. High-frequency data may be obtained from exchanges, consolidated feeds or commercial vendors. Issues such as missing records, out-of-sequence trades and corrections can affect bar statistics and often require preprocessing.
• Time zones and trading sessions. Time-bar construction requires consistent handling of time zones, daylight-saving changes and partial trading days. Many applications limit bar construction to regular trading hours and treat extended hours separately.
• Corporate actions. For equities, stock splits, dividends and changes in share count can alter price and volume series. These actions are often adjusted for when constructing long-span bar data sets, particularly for dollar bars.
• Auction and block trades. Opening and closing auctions, as well as large block trades, may generate trades with unusually large volume or notional value. Some applications treat these events separately to avoid large single trades dominating a bar.
• Multi-asset portfolios. When bar-constructed series are used for portfolio-level analysis, decisions must be made about synchronising bars across instruments, for example by using a common time grid or by intersecting periods where data are available.

These factors influence the properties of the resulting data set and are often documented in empirical studies that rely on bar-constructed series.

Applications

Bar-construction methods are used in a range of financial applications, including:

• estimation of intraday volatility and correlation;
• analysis of market microstructure, including order-flow and price-impact studies;
• construction of features for high-frequency trading strategies;
• back-testing of algorithmic execution and market-making models;
• risk management systems that rely on intraday data.

In many of these settings, researchers select a bar type and threshold so that the resulting series balances statistical properties, computational efficiency and interpretability for the task at hand.

See also

• Financial time series
• High-frequency trading
• Market microstructure
• Time series
• Volatility (finance)

References

Category:Financial data analysis Category:Econometrics Category:Algorithmic trading