Current-Relevance Risk Modeling
Model today's exposures with recent data and factor-based historical reconstruction
- Difficulty
- Advanced
- Time to result
- ~weeks to results
- Steps
- 6
- Confidence
- 97%
Long data sets appear statistically robust, but they can combine fundamentally different versions of an asset class. Sector weights, credit quality, duration, classifications, and underlying business models evolve, so an old index history may describe an exposure that no longer exists. Begin by defining the current asset through its sector and factor composition. Give recent, higher-frequency observations appropriate weight when forecasting near-term risk. Then extend the sample by backfilling the histories of today's factors rather than blindly using the original index return series. Where regimes matter, compare the present with historically similar environments. The output is a risk estimate grounded in current exposures while still extracting useful information from history.
Origin
Extracted from Masters in Business, where Sébastien Page challenged the assumption that the longest available investment data set is always the most useful.
Core principles
- 01More historical data is not automatically better data
- 02Model the asset that exists today, not an obsolete index composition
- 03Prefer recent observations when forecasting near-term risk
- 04Use factor histories to reconstruct current exposures
- 05Sample comparable regimes when structural conditions matter
How to run it
- 1
Define the forecast
Specify whether the model must estimate near-term volatility, long-term strategic risk, tail loss, or another quantity. Match the data frequency and history to that purpose.
Pro tip Use a shorter horizon for rapidly changing risk and a broader scenario set for strategic planning.
Watch out One data window cannot answer every risk question.
- 2
Audit structural change
Compare the asset's present composition with its historical composition. Examine sector weights, duration, credit quality, geography, classification changes, and dominant companies.
Pro tip Plot major exposure changes through time before fitting the model.
Watch out An unchanged index name does not imply an unchanged economic exposure.
- 3
Map current factors
Represent the asset using its present factor exposures, such as technology, duration, quality, cyclicality, or emerging-market platform exposure. Identify the factors most likely to drive current risk.
Pro tip Use economically interpretable factors that can be monitored over time.
Watch out An overly complex factor map can create unstable estimates.
- 4
Use recent data for current risk
Give recent, higher-frequency observations meaningful weight when estimating forward risk. Test whether they predict current volatility better than a full-history average.
Pro tip Compare several rolling windows rather than selecting one retrospectively.
Watch out Very short samples can overfit temporary shocks.
- 5
Backfill factor histories
Construct a longer synthetic history from the factors matching today's asset composition. Use this to examine how the current exposure might have behaved in earlier environments.
Pro tip Validate the reconstructed series against periods where both actual and factor data exist.
Watch out A backfill is a model, not an observed track record.
- 6
Apply regime judgment
Identify historical regimes that resemble current inflation, rate, growth, and liquidity conditions. Compare regime-specific estimates with the broader factor-based result and disclose uncertainty.
Pro tip Report a range of risk estimates instead of one falsely precise number.
Watch out No historical regime will perfectly reproduce the present.
In the wild
Page noted that technology's weight in the S&P 500 changed dramatically around the dot-com era and afterward. Using very old index data without adjustment therefore mixes portfolios with materially different sector exposures. A current-relevance model would map today's sector factors and reconstruct their earlier behavior.
→ The risk estimate better represents the modern S&P 500 rather than an obsolete sector mix.
The duration and quality composition of broad bond indexes changed over time. A historical average based on a shorter-duration index can understate the rate sensitivity of the current index. Modeling today's duration and credit factors separately makes the comparison more relevant.
→ The forecast reflects the current bond index's greater interest-rate exposure.
Common mistakes
Maximizing sample length blindly
A longer sample can add observations from an asset composition that no longer exists. Relevance must take priority over raw observation count.
Ignoring index drift
Sector classifications, constituent weights, credit quality, and duration can all change while the index retains the same label. Audit these changes explicitly.
Treating backfills as facts
Factor reconstructions depend on modeling assumptions and should be validated and labeled accordingly. They supplement observed data rather than replacing it without qualification.
Is it for you?
Best for
It is best for quantitative analysts, asset allocators, and advisers constructing forward-looking risk models from changing indexes.
Not ideal for
It is not ideal when current exposures cannot be measured or when the available factor histories are too sparse to support reconstruction.
From the transcript
“higher frequency, shorter term data are more predictive of risk going forward than the longer term data”
“you're basically modeling risk. The risk of an asset class that no longer exists”
“is more data always better than than more recent or more relevant data? The answers know.”
From the episode
Sébastien Page on Smart Asset Allocation
Sébastien Page