Skip to main content
Track Monte Carlo simulation
22 / 26
Search the archiveLibrary
Library

2015issue C0526-28

Constructing next-session implied-volatility forecasts by reordering closes

A fixed implied-volatility-index lookback can be ranked by close instead of by date so next-session-change values accumulate along a cumulative-delta path. The live overlapping-window cannot observe the trough-close, so mean-substitution or linear-regression supplies the split and the finished rule is judged against a monte-carlo-simulation of random direction calls.

  • A close-ranked-reorder treats calendar order as optional: the same implied-volatility-index window is sorted by descending close so next-session-change values can be accumulated out of calendar sequence.
  • The trough-close is the ranked close where the cumulative-delta path is lowest, and it is meant to split sessions that more often precede a decline from those that more often precede a rise.
  • A live overlapping-window cannot compute that trough-close, so mean-substitution or linear-regression supplies the missing split from already-known window inputs.
  • Monte-carlo-simulation of random up-or-down calls on the same sessions is used only as a construction baseline for the finished rule.
Entries in this reading3 entries
Current trackMonte Carlo simulation51 readingsLinear regression135 readingsImplied volatility analysis176 readings

Calendar order as a modeling choice

The archive stores an implied-volatility-index overlapping-window as the session date, the close, and the next-session-change assigned to the current row. Those three columns stay together when the window is reordered. The implied-volatility-index is the ordered input for the construction. It is not treated here as a standalone market-timing story.

Editorial reading: calendar order is a modeling choice. The construction inspects relationships that are not calendar-sequential by ranking the same lookback by close rather than by date.

Ranking closes and accumulating changes

A close-ranked-reorder sorts the stored rows by descending close. The next-session-change values are then accumulated in that non-calendar order to form a cumulative-delta path.

In the stored sample that path reaches a low and then turns higher. The close on the ranked row at that low is the trough-close, and it is the intended high-versus-low split.

Closes at or above that trough-close more often precede a next-session decline. Closes below it more often precede a next-session rise. In other overlapping windows the trough-close is frequently off-center rather than at the midpoint of the ranked series.

Cumulative next-session VIX change after ranking 30 closes

Rank the 30 daily VIX closes from 16 December 1996 through 28 January 1997 from highest to lowest and the attached next-session changes accumulate down to -6.84 at rank 15 (close 19.42 on 18 December 1996), then turn higher. That trough is the split: sessions at or above it more often fall next day; sessions below it more often rise. Every point is taken from the source table of ranked closes, next-session changes, and their running sum.
Rank the 30 daily VIX closes from 16 December 1996 through 28 January 1997 from highest to lowest and the attached next-session changes accumulate down to -6.84 at rank 15 (close 19.42 on 18 December 1996), then turn higher. That trough is the split: sessions at or above it more often fall next day; sessions below it more often rise. Every point is taken from the source table of ranked closes, next-session changes, and their running sum.CBOE Volatility Index (VIX) · daily · 1996-12-16T00:00:00.000Z to 1997-01-28T00:00:00.000Z

Next-session change is the following trading day's move in calendar time, carried with the row when the window is re-ranked; it is not the step to the next row after sorting. The 28 January 1997 row uses the subsequent session outside the 30-day window.

Why the live trough-close is missing

A window that ends on the current session lacks the final next-session-change. The live ranked series therefore cannot compute the cumulative-delta trough, and the trough-close must be estimated before a same-session forecast can be formed.

Estimating the unobserved split

Historical trough-close values tend to cluster near the middle of each ranked window. Mean-substitution therefore replaces the unobserved trough-close with the already-known window mean or median close.

Linear-regression is offered as an alternative estimator. A least-squares fit on several known window inputs is used to estimate the missing trough-close when the latest next-session-change is not yet available.

A direction null for the finished rule

The finished construction is examined on overlapping windows rebuilt at each new close. A monte-carlo-simulation of random daily up-or-down calls on those same sessions is used only as a construction baseline for the finished rule.

Educational research material, not investment advice. Historical source context does not establish present-day performance.
22 of 26 in the Monte Carlo simulation track
201632-37 pp.Next on Monte Carlo simulationMechanical systems as a gated factoryMechanical trading remains emotional; the feeling attaches to following or ignoring a written rule set.
All readings on this track · 26 readings
  1. 1989Known-bias price streams for system evaluation
  2. 1989Constructing holdout-checked synthetic prices from chaos maps
  3. 1991Closing the feedback loop in auction market simulations
  4. 1994Evaluating eight-year election regimes
  5. 1995Constructing equity versus leverage with resampled trades
  6. 1995Stock leverage from compounded returns and margin cost
  7. 1996Constructing and optimizing volatility position-sizing profiles
  8. 1996Constructing fade rules from an empirical deviation histogram
  9. 1998Return-path volatility and the recovery burden
  10. 1999Monte Carlo checks beyond a single historical path
  11. 2002Monte Carlo equity paths from win rate and profit factor
  12. 2003Checking a small system edge with Monte Carlo simulation
  13. 2003Evaluate a mechanical trading system with a stop-loss order and Monte Carlo simulation
  14. 2005Evaluating momentum rank rotation under portfolio constraints
  15. 2009Bounded risk and rank rotation in portfolio construction
  16. 2010A calendar for walk-forward robustness and Monte Carlo evaluation
  17. 2012Treat the optimize view as an evaluation protocol
  18. 2013Entry, exit and abstention as one testable procedure
  19. 2014Evaluating stop and target bounds before fading large moves
  20. 2014Evaluating system exits when markets get noisier
  21. 2015Memory-bearing synthetics for dominant-cycle spectra
  22. 2015Constructing next-session implied-volatility forecasts by reordering closes
  23. 2016Mechanical systems as a gated factory
  24. 2017Four-sleeve Monte Carlo annual-loss bounds
  25. 2018When to retire a mechanical system with Monte Carlo drawdown limits
  26. 2020Treat mechanical system design as a revalidation lifecycle
By fieldTime series and forecasting58 readingsBy year2015181 readings
All 51 readings tagged Monte Carlo simulation
Also on Monte Carlo simulation5 readings
1984When random walks fail to explain market averages1986Evaluating advance-decline plus-day runs against chance baselines1989Constructing holdout-checked synthetic prices from chaos maps1989Known-bias price streams for system evaluation1989Measured range, scale-in ladders, and spread replay as a cash-reserve design