The riskiest part of this strategy would definitely be UVXY, which is 1.5x the VIX. Replacing it with SQQQ (3x inverse Nasdaq 100) would reduce the volatility but also reduce returns.
Above the 200-day moving average (bull market), TQQQ (3x Nasdaq-100) is generally a good idea, but be prepared for the value to fluctuate 3.5x compared with the S&P. For example, if the S&P is down 1.5%, TQQQ would be down more than 5%. If that scares you, then you could switch to QLD (2x Nasdaq-100).
Below the 200-day moving average (bear market), the risk would of course be exactly the same as the S&P since the strategy holds SPY.
Yeah, I actually really liked the TQQQ element, seems very simple yet completely logical - buy when market is trending up and sell when overbought. I feel like often enough simple works best (unless you’re a quant).
But as you’ve said, VIX was the bet I was curious about, might consider as you’ve suggested and simply short the market rather than long VIX to reduce beta.
Thanks for the quick response - super informative! Will be looking out for your future posts :)
Excellent article. Thanks for this fascinating illustration. Did you consider generating synthetic TQQQ prices for dates before 2011-02-09 (TQQQ inception) and after 1999-03-10 (QQQ inception)? Obviously, I'm thinking that the 2000-02 and 2008-09 periods would both have been very difficult for any leverage rotation strategy with a long duration trend filter, like the 200 DSMA. On the use of a long VIX ETP, here UVXY, am I right to take the statement that 41.8% of total net profit came from the 3% of time exposed to this instrument to mean that, of the total cumulative strategy return of 158,022% over the whole 13.5 years, some 66,053% of it came from this limb (such that, had the strategy instead sat in cash for the 3% of time that it was long UVXY, and assuming no return on cash, then the strategy would have still returned, without UVXY exposure, 91,968% overall?) Thank you in advance for your clarifications, and thank you again for the engaging and well written article.
Thank you for the close read and thoughtful questions.
First, we did not generate synthetic TQQQ prices for the pre-inception period. We deliberately started the backtest in October 2011 to rely entirely on actual ETF data. You are correct that 2000-2002 and 2008-2009 would have been exceptionally difficult for this strategy due to violent whipsaws around the 200-day moving average. However, synthesizing a 3x leveraged series often fails to capture true daily rebalancing drag and tracking error, making those historical periods look artificially clean on paper. Sticking to live data keeps the metrics grounded in reality.
Second, regarding the UVXY math: the 41.8% figure represents a share of total net profit in absolute dollars, not a slice of the cumulative percentage return. Linearly subtracting a profit slice from a geometric return does not work mathematically. Removing those specific trades alters the equity curve and resets the compounding capital base for every subsequent position. The return is meaningfully lower without those volatility spikes, but it is not a straight subtraction down to 91,968%.
Finally, it's important to note that this strategy is shared purely for demonstration purposes. It is far too volatile to command anything more than a small, speculative allocation in a broader portfolio.
Thanks for the super article. Couple of questions if I may. Did you use the Simple or Exponential 200 DMA for the S&P 500? Are you using the close level on the S&P 500 (rather than intraday/in session) for the 200 DMA signal? Did you consider tolerance thresholds around the 200 DSMA (e.g. + or - 3% or 4%) to reduce whipsaw trading into and out of TQQQ and SPY? Sorry to ask so many questions, and thanks in advance for your answers. Really enjoyed and appreciated your article.
- Tolerance thresholds are not included in this algo for simplicity's sake, but they can be used.
On tolerance thresholds specifically, there are trade-offs to consider.
On the upside, adding a buffer zone (e.g. +/- 3-4% around the 200 DMA) can reduce whipsaw trades during periods when price is oscillating around the moving average. This means fewer false signals, lower transaction costs, and less slippage from unnecessary round trips. It can also improve signal quality by filtering out noise and only acting on more decisive trend changes.
On the downside, tolerance bands introduce a lag. You're effectively waiting for a larger move before acting, which means delayed entries and exits. In a sharp selloff or rally, that extra 3-4% buffer could erode returns. There's also the issue of parameter sensitivity: the "right" threshold isn't static and what works well in one regime may underperform in another.
Not investment advice, of course. Do your own research.
Thank you for an excellent piece. Thinking in comparison with Michael Gayed's Leverage for the Long Run UPRO 200 DSMA SPY/S&P 500 Leverage Rotation Strategy, why chose SPY as the risk off asset, and not, say, T-Bills or cash (or, for that matter, Intermediate or Long duration Treasuries, or even those and some Gold)? Did SPY give better CAGR and/or Sharpe/Sortino overall? Thanks again for your thought provoking piece.
You're welcome. Yes, SPY produced a better CAGR overall. Snap-back rallies during bear markets can be vicious, and some of the strongest up-days in history have occurred in the midst of drawdowns. By rotating into SPY rather than cash or Treasuries, the strategy captures a portion of that upside without taking on the full downside of leveraged positions.
The "Triple Accelerator" algorithm is an innovative strategy that adjusts investments based on market trends, but it's important to be mindful of the risks involved, including potential drawdowns. While the results are impressive, careful consideration and understanding are key.
Is there anything you’d add to this strategy to improve its accuracy?
It's a decent but volatile strategy as it is.
The riskiest part of this strategy would definitely be UVXY, which is 1.5x the VIX. Replacing it with SQQQ (3x inverse Nasdaq 100) would reduce the volatility but also reduce returns.
Above the 200-day moving average (bull market), TQQQ (3x Nasdaq-100) is generally a good idea, but be prepared for the value to fluctuate 3.5x compared with the S&P. For example, if the S&P is down 1.5%, TQQQ would be down more than 5%. If that scares you, then you could switch to QLD (2x Nasdaq-100).
Below the 200-day moving average (bear market), the risk would of course be exactly the same as the S&P since the strategy holds SPY.
Yeah, I actually really liked the TQQQ element, seems very simple yet completely logical - buy when market is trending up and sell when overbought. I feel like often enough simple works best (unless you’re a quant).
But as you’ve said, VIX was the bet I was curious about, might consider as you’ve suggested and simply short the market rather than long VIX to reduce beta.
Thanks for the quick response - super informative! Will be looking out for your future posts :)
You're welcome! Glad you liked it.
Excellent article. Thanks for this fascinating illustration. Did you consider generating synthetic TQQQ prices for dates before 2011-02-09 (TQQQ inception) and after 1999-03-10 (QQQ inception)? Obviously, I'm thinking that the 2000-02 and 2008-09 periods would both have been very difficult for any leverage rotation strategy with a long duration trend filter, like the 200 DSMA. On the use of a long VIX ETP, here UVXY, am I right to take the statement that 41.8% of total net profit came from the 3% of time exposed to this instrument to mean that, of the total cumulative strategy return of 158,022% over the whole 13.5 years, some 66,053% of it came from this limb (such that, had the strategy instead sat in cash for the 3% of time that it was long UVXY, and assuming no return on cash, then the strategy would have still returned, without UVXY exposure, 91,968% overall?) Thank you in advance for your clarifications, and thank you again for the engaging and well written article.
Thank you for the close read and thoughtful questions.
First, we did not generate synthetic TQQQ prices for the pre-inception period. We deliberately started the backtest in October 2011 to rely entirely on actual ETF data. You are correct that 2000-2002 and 2008-2009 would have been exceptionally difficult for this strategy due to violent whipsaws around the 200-day moving average. However, synthesizing a 3x leveraged series often fails to capture true daily rebalancing drag and tracking error, making those historical periods look artificially clean on paper. Sticking to live data keeps the metrics grounded in reality.
Second, regarding the UVXY math: the 41.8% figure represents a share of total net profit in absolute dollars, not a slice of the cumulative percentage return. Linearly subtracting a profit slice from a geometric return does not work mathematically. Removing those specific trades alters the equity curve and resets the compounding capital base for every subsequent position. The return is meaningfully lower without those volatility spikes, but it is not a straight subtraction down to 91,968%.
Finally, it's important to note that this strategy is shared purely for demonstration purposes. It is far too volatile to command anything more than a small, speculative allocation in a broader portfolio.
Thanks for the super article. Couple of questions if I may. Did you use the Simple or Exponential 200 DMA for the S&P 500? Are you using the close level on the S&P 500 (rather than intraday/in session) for the 200 DMA signal? Did you consider tolerance thresholds around the 200 DSMA (e.g. + or - 3% or 4%) to reduce whipsaw trading into and out of TQQQ and SPY? Sorry to ask so many questions, and thanks in advance for your answers. Really enjoyed and appreciated your article.
Glad you enjoyed the article.
Happy to answer your questions:
- Simple 200 DMA for the S&P
- Yes, close level on the S&P
- Tolerance thresholds are not included in this algo for simplicity's sake, but they can be used.
On tolerance thresholds specifically, there are trade-offs to consider.
On the upside, adding a buffer zone (e.g. +/- 3-4% around the 200 DMA) can reduce whipsaw trades during periods when price is oscillating around the moving average. This means fewer false signals, lower transaction costs, and less slippage from unnecessary round trips. It can also improve signal quality by filtering out noise and only acting on more decisive trend changes.
On the downside, tolerance bands introduce a lag. You're effectively waiting for a larger move before acting, which means delayed entries and exits. In a sharp selloff or rally, that extra 3-4% buffer could erode returns. There's also the issue of parameter sensitivity: the "right" threshold isn't static and what works well in one regime may underperform in another.
Not investment advice, of course. Do your own research.
Thank you for an excellent piece. Thinking in comparison with Michael Gayed's Leverage for the Long Run UPRO 200 DSMA SPY/S&P 500 Leverage Rotation Strategy, why chose SPY as the risk off asset, and not, say, T-Bills or cash (or, for that matter, Intermediate or Long duration Treasuries, or even those and some Gold)? Did SPY give better CAGR and/or Sharpe/Sortino overall? Thanks again for your thought provoking piece.
You're welcome. Yes, SPY produced a better CAGR overall. Snap-back rallies during bear markets can be vicious, and some of the strongest up-days in history have occurred in the midst of drawdowns. By rotating into SPY rather than cash or Treasuries, the strategy captures a portion of that upside without taking on the full downside of leveraged positions.
The "Triple Accelerator" algorithm is an innovative strategy that adjusts investments based on market trends, but it's important to be mindful of the risks involved, including potential drawdowns. While the results are impressive, careful consideration and understanding are key.
That's correct