A single correlation coefficient is a summary statistic. Like all summary statistics, it compresses a complex, time-varying relationship into a single number, and in doing so it discards the information that matters most for understanding how that relationship actually behaves. The overall correlation between the TTU TF Index and the S&P500TR Index across the full period from January 2000 to the present is approximately -0.12. This figure suggests near-zero correlation and would, taken at face value, classify the two as effectively uncorrelated. The same calculation for the Vanguard Total Bond Market Index Fund (VBMFX) against the S&P500TR produces a correlation of 0.13, similarly near zero. On the basis of this single statistic, both appear to be equivalent diversifiers for a traditional US equity portfolio.
They are not equivalent. And the single statistic is not the right tool for determining that.
The Misuse of Single Statistics
The correlation between trend-following strategies and equity indices is not static. It is regime-dependent: the relationship switches character depending on the market environment in which it is observed. During periods of market expansion, trend-following strategies tend to participate in the upward momentum, producing positive correlation with equities. During market dislocations, when equities are falling sharply, trend-following strategies adapt to the directional change and frequently produce negative correlation with equities, generating returns precisely when the equity portfolio is suffering its largest drawdowns.
The aggregate correlation of -0.12 is the mathematical average of these two distinct regimes across the full twenty-four-year period. It does not describe the relationship at any specific point in time. It describes the net result of a relationship that has been systematically positive during booms and systematically negative during busts. These two properties offset each other in the aggregate statistic, producing an apparent neutrality that conceals the directionally valuable structure underneath.
This is the central problem with relying on single statistics to evaluate diversification: they average across regimes rather than revealing the regime-dependent structure that determines whether an asset actually provides protection when protection is needed.
Figure 1: Single Statistic Correlations between TTU TF Index, VBMFX against S&P500TR — covering the period 1 January 2000 to 31 May 2024
Visual Correlation Mapping
Visual correlation mapping provides a more informative representation of the relationship between assets by tracking how their price movements align or diverge across the full time series, rather than collapsing the relationship to a single number.
Figure 2 maps the correlation between the S&P500TR Index and the TTU TF Index since January 2000. Blue boxes mark periods of positive correlation, where both series move in the same direction. Red boxes mark periods of negative correlation, where they move in opposite directions. The switching between these two states is clearly visible and maps directly onto recognisable market regimes: the periods of positive correlation align with market expansions, while the periods of negative correlation align with the major market dislocations of 2001 to 2003, 2008 to 2009, and 2020.
Figure 2: Visual Correlation Mapping Between S&P500TR and the TTU TF Index since 1st Jan 2000
Figure 3 maps the same relationship for the S&P500TR Index and the VBMFX. The historical pattern shows a similar regime-dependent structure across most of the period: negative correlation during equity market stress, as capital moved from equities into bonds in a flight to safety, and positive correlation during periods of general market stability. However, since approximately 2020, this historical pattern has broken down materially. The equity-bond correlation has shifted to a persistently positive relationship, with both asset classes moving in the same direction. The diversification benefit that the 60/40 portfolio historically relied upon has deteriorated precisely as the inflationary regime of 2022 demonstrated: equities and bonds fell simultaneously, providing no offsetting protection.
Figure 3: Visual Correlation Mapping Between S&P500TR and the VBMFX since 1st Jan 2000
Rolling Correlation Analysis
Rolling correlation analysis extends the visual mapping by quantifying how the correlation between assets evolves across different time windows, providing a more precise picture of the regime-dependent dynamics that the single statistic obscures.
Figure 4 shows the 12-month rolling correlation between the S&P500TR and the TTU TF Index from 2000 to 2024. The correlation fluctuates substantially across the period, ranging from deeply negative during the major equity market dislocations to moderately positive during sustained equity expansions. This range of variation confirms what the visual mapping indicated: the relationship is genuinely regime-dependent and retains that property consistently across the full history. The TTU TF Index has maintained its character as a strategy that adapts to prevailing market conditions, producing correlation that is directionally aligned with equity markets during expansions and inversely aligned during dislocations.
Figure 4: 12-Month Rolling Correlation Between S&P500TR and TTU TF Index
Figure 5 shows the 12-month rolling correlation between the S&P500TR and VBMFX across the same period. For most of the history, the relationship oscillates between positive and negative, broadly reflecting the traditional flight-to-safety dynamic. The structural break from approximately 2020 onward is clearly visible: the rolling correlation shifts to a persistently elevated positive level, reflecting the regime shift in which bonds and equities became jointly sensitive to the inflationary environment. The diversification property that made the 40% bond allocation in a 60/40 portfolio functionally valuable has not merely weakened. It has, in the current regime, reversed.
Figure 5: 12-Month Rolling Correlation Between S&P500TR and VBMFX
Regime Transition Analysis
Regime transition matrices provide a quantitative framework for understanding not only the current correlation regime between two assets but the probability of transitioning from one regime to another. The following matrices classify the S&P500TR relationship with both the TTU TF Index and the VBMFX into three regimes: uncorrelated, positively correlated, and negatively correlated.
Figure 6: Regime Transition Matrix Between S&P500TR and TTU TF Index
Figure 7: Regime Transition Matrix Between S&P500TR and VBMFX Index
For the S&P500TR and TTU TF Index pair, the uncorrelated regime is stable, with a probability of 0.889 of remaining in the same regime. Once the relationship enters the positively correlated regime, it is highly persistent, with a probability of 0.951 of remaining there. The negatively correlated regime is also strongly persistent at 0.900. The transition probabilities between regimes are low, reflecting that each correlation regime, once established, tends to persist across a meaningful period. This persistence is the property that makes the regime-dependent correlation of the TTU TF Index valuable as a diversifier: when the negative correlation regime arrives, during equity market stress, it is likely to persist for long enough to provide sustained protection rather than a brief and unreliable offset.
For the S&P500TR and VBMFX pair, the uncorrelated regime is similarly stable at 0.904. The positively correlated regime is somewhat less persistent at 0.839, and the negatively correlated regime is materially less persistent at 0.735, with a transition probability of 0.265 from the negatively correlated state back to the uncorrelated state. The bond relationship is structurally less stable within its correlation regimes than the trend-following relationship. The diversification it offers during equity market stress is real but less durable: the negative correlation regime is more likely to break down and transition to uncorrelated before the equity market stress has fully resolved.
The comparison confirms what the visual and rolling analyses indicated. The TTU TF Index maintains more stable, more persistent correlation regimes relative to the S&P500TR than the VBMFX does. And since 2020, the VBMFX has shifted into a regime in which the negative correlation that historically defined its diversification value has been largely absent.
Lifting Power and Portfolio Geometry
Correlation properties alone are insufficient to evaluate the contribution of an asset to a diversified portfolio. The other dimension is lifting power: the extent to which adding the asset improves the portfolio’s geometric return outcomes rather than merely reducing its volatility. An asset that provides negative correlation during equity market stress but produces negligible returns across the full cycle adds protection without adding compounding. The relevant question is whether the asset improves both the drawdown characteristics and the CAGR of the combined portfolio.
Figure 8 compares portfolio geometry across five allocations: the S&P500TR alone, the TTU TF Index alone, the VBMFX alone, a 60% S&P500TR / 40% TTU TF Index blend, and a 40% S&P500TR / 60% TTU TF Index blend, alongside the traditional 60% Equity / 40% Bond portfolio. The Sharpe and Sortino ratios are included as conventional reference points only. The Sharpe Ratio penalises beneficial volatility, which means it systematically understates the quality of positively skewed strategies. The primary performance measures are CAGR, maximum drawdown, and MAR.
Figure 8: Comparative Portfolio Geometry Metrics across allocations — S&P500TR, TTU TF Index, VBMFX, and blended portfolios, covering 1 January 2000 to 31 May 2024
The results are unambiguous. Both the 60/40 and 40/60 equity-to-trend-following blends produce materially higher CAGR, lower maximum drawdown, and better MAR ratios than the traditional 60% Equity / 40% Bond portfolio across the full twenty-four-year period. The trend-following allocations deliver superior geometric return outcomes and more favourable drawdown characteristics across every allocation tested. The optimal allocation, on the basis of this analysis, is not the conventional 40% bond / 60% equity split but a 60% trend-following / 40% equity split, which produces the strongest combination of CAGR and drawdown management across the full cycle.
The TTU TF Index generates lifting power, not merely correlation protection. It improves the geometric return outcomes of the combined portfolio in addition to providing the regime-dependent negative correlation that protects the equity component during market dislocations. The VBMFX, by contrast, has historically provided modest lifting power, with its 3.65% CAGR contributing limited compounding benefit relative to its 18.53% maximum drawdown, and since 2020 its correlation protection has deteriorated structurally.
The Case for a Material Trend Following Allocation
The twenty-four-year record examined here makes a specific and quantifiable argument. Trend following is not merely a hedge against equity market stress. It is a superior diversifier that improves portfolio geometry across the full market cycle, combining regime-dependent negative correlation during equity dislocations with meaningful positive lifting power during expansionary periods. The diversification benefit it provides is structural, arising from the inherent flexibility of a strategy that can position for both rising and falling prices, rather than from a fixed historical correlation that is contingent on a specific macroeconomic regime remaining stable.
The equity-bond correlation breakdown that has unfolded since 2020 has made this argument more urgent, but the data shows it was already compelling across the full history. A material allocation to trend following has consistently produced better portfolio geometry than the conventional bond allocation, not because market conditions recently turned unfavourable for bonds, but because trend following’s regime-adaptive properties have always made it the more structurally sound diversifier. The evidence has been present in the data throughout. The single statistic just did not reveal it.