
Event Clocks from two venues: separating common news from venue noise
Source:R/measure.R
event_clock_venues.RdWhen the same event trades on two venues (say Polymarket and Kalshi), each venue's measured clock contains the common information flow plus venue-specific noise (bid-ask bounce, stale quotes, idiosyncratic flows). Comparing four clocks built from synchronized prices separates the two:
A_x,A_yeach venue's own clock, \(\sum (\Delta L)^2\).
A_compositethe clock of the equal-weight average of the two venues' log odds.
A_crossthe cross-venue clock \(\sum \Delta L^x \Delta L^y\): news moves both venues, so it survives the product; independent venue noise does not, so on average it drops out.
Arguments
- x, y
event_pricesobjects (or coercible) for the same event outcome on two venues.- from
Valuation date/time (default: first observation).
- to
One or more horizon dates/times (default: last observation = full sample). Names are used as horizon labels.
- labels
Optional venue labels (default: the
market_ids).- clip
Numeric length-2 clipping bounds for
qbefore the log-odds transform; defaults to the bounds stored inx.
Value
A tibble with one row per horizon: venue_x, venue_y,
from, to, horizon, n_common (synchronized observations),
A_x, A_y, A_composite, A_cross, corr (correlation of the two
venues' log-odds changes), noise_share_x, noise_share_y.
Details
Under a simple benchmark in which each venue's log odds are the common
latent log odds plus independent noise, \(E[A_x] = A + N_x\) while
\(E[A_{cross}] = A\). The reported noise_share_x
\(= 1 - A_{cross}/A_x\) is then the fraction of venue x's measured
clock attributable to its own noise. In finite samples A_cross can
fall below zero when there is little common news; it is a diagnostic,
not an exact measurement. The identity
\(A_{composite} = (A_x + A_y + 2A_{cross})/4\) holds exactly.
Multiple venues help diagnose timing and filter venue-specific noise,
but they cannot by themselves pin down the absolute physical clock
scale; that still rests on the recovery restriction (see
ec_recovery_set()).
Prices are matched on identical timestamps. Synchronize intraday series
first, e.g. with a common daily snapshot:
pm_daily(x, snapshot_hour = 15.75) takes the last price at or before
15:45 New York time.
References
Hanke, Schadner, Stöckl, and Weissensteiner (Working Paper), "The Event Clock: What State Prices Identify About Physical Information".
See also
event_clock_venues_path() for the cumulative clocks.
Examples
# two noisy venues observing the same simulated news
set.seed(7)
p <- ec_simulate_path(1, 120, q = 0.5, A = 1)
days <- as.Date("2024-06-01") + 0:120
noisy <- function(sd) data.frame(time = days,
q = ec_ilogit(p$L + stats::rnorm(121, sd = sd)))
event_clock_venues(noisy(0.05), noisy(0.10), labels = c("A", "B"))
#> # A tibble: 1 × 13
#> venue_x venue_y from to horizon n_common A_x A_y A_composite
#> <chr> <chr> <date> <date> <chr> <int> <dbl> <dbl> <dbl>
#> 1 A B 2024-06-01 2024-09-29 2024-0… 121 1.32 2.94 1.50
#> # ℹ 4 more variables: A_cross <dbl>, corr <dbl>, noise_share_x <dbl>,
#> # noise_share_y <dbl>