Bitcoin trading strategy using SignalSolver Sentiment

Bitcoin Trading Strategy using SignalSolver Sentiment

This trading strategy for Bitcoin uses SignalSolver Sentiment to generate trading signals. It is currently showing a 22% drawdown, but it has been a good performer overall. Sentiment on any given day is determined by backtesting up to that day, selecting the top algorithms and evaluating the ratio of long to short positions. Sentiment (percentage bullish) is shown as a blue bar-chart on the graph. Trading is performed subsequent to the Sentiment crossing a threshold value.

The Buy and Sell thresholds for this simulation (green and red lines on the time graph) used an adaptive technique every cycle to determine the optimum values, with the constraints Buy+Sell=100% and Buy>Sell. The optimum thresholds settled down pretty quickly to Buy=55% and Sell=45%. Return was $67,966 since 1/1/21 for a $10,000 outlay (CAGR 0f 273%), with Risk-Reward about 680 times that for Short-hold. The simulations assume the ability to both long and short the security, but the table includes both the Long and Short returns

Bitcoin trading using SignalSolver Sentiment with adaptive symmetric thresholds

Bitcoin trading using SignalSolver Sentiment with adaptive symmetric thresholds

In selecting a threshold strategy, you couldn't get into too much trouble since the other two common threshold strategies also showed a profit. Using an adaptive threshold with the Buy=Sell constraint applied returned $25,206 (123% CAGR)--

Bitcoin trading using SignalSolver Sentiment, adaptive equal thresholds

Bitcoin trading using SignalSolver Sentiment, adaptive equal thresholds

Finally applying a simple constant 50% threshold also worked well, returning $40,640 (183% CAGR).  From this image you can view the entire threshold surface, which is  what we like to see--a solid looking structure.

Bitcoin trading using SignalSolver Sentiment with a fixed 50% threshold

Bitcoin trading using SignalSolver Sentiment with a fixed 50% threshold

Settings are shown below:

SignalSolver settings for the Bitcoin Sentiment runs

SignalSolver settings for the Bitcoin Sentiment runs

These are the same settings we used for TQQQ with the only change being that we (somewhat arbitrarily) increased the number of algorithms to 21 (from 5 for TQQQ).  Only OHLC prices and short period (10 day) SMAs and EMAs are used in the backtests.

As with all Sentiment runs, out-of-sample walk-forward simulation is used. Sentiment is determined for a particular day by using the backtester on prior data, in this case using the ratio of the top 21 algorithms' long (bullish) to short (bearish) state. When the sentiment crosses the threshold in either direction, a trade is performed appropriately. In the case of Bitcoin, OHLC prices run from midnight to midnight.

TQQQ

AAPL signals prove versatile

AAPL signals prove versatile

Work on TQQQ, FNGU, NFLX and many others

 

The SignalSolver Sentiment indicator is calculated by aggregating multiple algorithm sentiments. The Sentiment technical indicator has a value between 0% (completely bearish) to 100% (completely bullish). Intuitively, you would think that a 50% threshold would be the best threshold value from which to generate buy and sell signals, and often this is the case. However, in the case of AAPL, the overall best Buy=Sell threshold (for the settings we are using) has been 85%. The result is shown below in the "L&S" column of the table:

AAPL performance using an 85% threshold

AAPL performance using an 85% threshold

The buy/sell signals generated by this algorithm have shown surprising versatility. Below are shown images showing the result of using the AAPL signals for AMZN, FB, FNGU, GOOG, MSFT, NFLX, NVDA, QQQ, TECL and TQQQ. In each case the result was better than both buy-hold and short-hold, often by an enormous margin.

The AAPL 85% Threshold

Success at using Sentiment to signal trades comes down to finding the best threshold to use. There are always values for the buy and sell threshold that generate equal or better returns than both buy-hold and short-hold. But you can only know these values retrospectively.

There was no way to predict that an 85% threshold would be optimum. However we can simulate (or run live) using an adaptive thresholds where the program optimizes the thresholds daily as you go along. When using this method, (as you might do in reality) the Buy=Sell constrained threshold was in the 80-90% range from Sept 2020 to Aug 2021 but then became unstable, with declining returns.

AAPL signal performance using adaptive threshold with constraint Buy=Sell

AAPL signal performance using adaptive threshold with constraint Buy=Sell

The threshold surface (shown below) is a very nice solid structure yielding a positive return for all constant values of threshold except the extreme edges. Notably, 50/50 yielded a return of twice that of buy-hold. The Buy>Sell adaptive threshold (also shown below) gave a better return than the Buy=Sell adaptive threshold, Here we optimize the thresholds every cycle as before.

AAPL Result using adaptive threshold with Buy>Sell constraint

AAPL Result using adaptive threshold with Buy>Sell constraint

Comparing the two results, you can see that when the Buy=Sell constraint is applied, the adaptive threshold appears to oscillate between the optimum buy threshold region and the optimum sell threshold region.