Temporal Coverage Bias in Financial Panel Data: A Coverage-Aware Structuring Framework with Evidence from the Dhaka Stock Exchange

Abstract

A common practice in empirical finance is to construct calendar-aligned panels that implicitly treat all instruments as having existed for the full observation period. When securities with different listing histories are combined without explicit coverage constraints, price histories can be inadvertently extended before valid trading ever began. This paper formalizes this problem and proposes a coverage-aware structuring framework built around instrument-level observation windows encoded through structured metadata and an availability matrix. Applied to end-of-day data from the Dhaka Stock Exchange spanning October 2012 to January 2026 and covering 486 instruments, the framework reveals substantial distortions from naive temporal alignment. ARIMA-based experiments establish the mechanism through which padded observations corrupt return dynamics, and volatility analysis across 53 instruments shows that forward-filling alone suppresses return volatility by roughly 20% on average, with GARCH unconditional variance distortions exceeding 26% in over 90% of instruments - a lower bound, as backward extension to the panel start produces 36.6% suppression and causes GARCH non-convergence in 41% of instruments. The distortion affects any method requiring calendar alignment of heterogeneous histories, including dynamic time warping, covariance-based portfolio construction, factor model regression, and temporal foundation model fine-tuning. Although demonstrated on financial data, the framework applies to any panel combining entities with heterogeneous entry dates, including sensor networks, clinical cohorts, and country-level economic panels. Listing coverage is not a minor preprocessing detail but a first-order variable in panel construction.

Figures (7)

• Figure 1: Instrument composition of the DSE end-of-day dataset by asset class. The dataset comprises 486 instruments across six categories, dominated by equities (395) and treasury bills (55). The multi-asset composition contributes to the heterogeneity of listing intervals observed in the panel.
• Figure 2: Distribution of instrument lifespans in calendar days. The spike at approximately 5,400 days corresponds to instruments listed since the start of the dataset observation window. The substantial proportion of instruments with lifespans below 2,000 days highlights the degree of listing heterogeneity and the extent of artificial backward padding that naive calendar alignment would introduce.
• Figure 3: Number of available instruments per trading day across the dataset observation period (October 2012 -- January 2026). Orange indicates instruments available in both adjusted and unadjusted versions; blue indicates availability in at least one version. Instrument count grows from approximately 260 to 415, with panel coverage rising from 53% to 85% of the 486 total instruments. The drop in mid-2020 corresponds to the DSE market suspension during the COVID-19 pandemic. The concave dip around mid-2024 reflects market disruption during the July 2024 uprising in Bangladesh. Both are genuine market events preserved in the dataset without imputation.
• Figure 4: Rolling ARIMA(1,1,1) forecast illustration for SQURPHARMA under the coverage-aware construction. Blue: training data over the valid listing interval; orange: test data (overlaps closely with training data at this scale); green: rolling one-step-ahead forecast. SQURPHARMA predates the panel start so coverage bias does not affect this specific instrument; the figure illustrates the forecast methodology under a clean coverage-aware construction. For instruments listed after 2016, naive calendar alignment would pad the training window with zero-return observations on non-trading days and, under backward extension, on pre-listing dates as well, suppressing return variance and distorting model estimates.
• Figure 5: Empirical distribution of volatility distortion across 53 instruments under the naive forward-filled construction. Return volatility distortion values (blue) are concentrated near 20%, while GARCH unconditional variance distortions (orange) exhibit a wider distribution with a higher mean of 26.2%, confirming that naive temporal alignment systematically biases volatility estimation downward. Both distributions reflect forward-filling only; the backward-filled construction produced GARCH non-convergence in 41% of instruments, representing a more severe failure mode.