VORTECS™ Score Live Testing Report: Jan 5 – Feb 8 2021

Summary

  • The VORTECS™ score compares the current state of the market to historical states
    • High scores indicate that historically-similar market conditions have led to price increases in the past.
    • Low scores indicate that historically-similar market conditions have led to price decreases in the past.
    • The further the deviation from the average score (50), the more confidence the algorithm has that current conditions have similar historical precedents.
  • The score looks at historical 24-hour price changes after certain market conditions.
    • However, the average time to peak price amplitude (up or down) ranges from a few hours to several days. See testing results.

Key findings:

  • We tested strategies including:
    • hold tokens above a score of 80
    • Buy a token if it becomes above 80. Sell once it falls below 60.
  • As mentioned above, timing is secondary to assessing market outlook in terms of historical directional change and magnitude. So strategies with strict timing rules (e.g. buy at 80, sell in 24 hours) tend to do worse than score-based strategies.
    • However, timing-based strategies still do beat the market on average.

The VORTECS™ Score is a score ranging from 0 – 100 applied to each cryptocurrency over time.

This score represents the patterned behavior of a particular coin. In other words, it identifies patterns in four variables (sentiment, price, tweet volume and trade volume) that typically lead to consistent outcomes. High scores (70+) mean that the coin typically sees increases in price after patterns similar to the current “state” (current set of patterns, or shapes) of the coin.

This score can, of course, be used in trading strategies. Here, we look at a few simple strategies, as well as some basic statistics.

The Dataset

Statistics here are generated from actual scores spanning one month, January 5th – February 8th, 2021.

Of course, it would be ideal to have statistics generated from back-tested scores spanning more historical time (e.g., all of 2020). However, the TEMPS score is relatively computationally demanding, taking about 10 seconds to generate a single score per coin. Consequently, back-tests so far have been limited to a subset of coins, calculated at a frequency of about one score every 6 hours (4 per day).

This is much lower quality than the scores we have running in real-time, which generate 70-100* scores per day, per coin. In the future, we plan to improve our back-testing infrastructure to handle generating scores at the same resolution as those seen on Markets Pro, in a reasonable amount of time.

*Note these scores refer to the raw VORTECS™ score, which is not actually ever shown to the user. The user sees a smoothed score over the last 12 hours (updated every 5 minutes), which is thus an average of 35-50 raw scores (half of the last day of scores). This smoothing provides a more accurate estimation of market conditions.

Statistics

Here we show the average returns after scores cross certain thresholds (70, 80, and 90). Note that scores can stay above these thresholds for a while; here, we are starting the “timer” when it initially crosses above these scores. 

Day Change and Week Change are simply the % change in price of an asset from the time the asset’s score first crossed a threshold until 24 and 168 hours later respectively.

These values can of course be skewed by the overall market: if everything went up 5%, it is not that significant that a particular coin went up 5%.

To address this, we also consider dominance change (“Dom Change”). This is how much the coin’s market cap dominance changed over the same period. Market cap dominance is the market cap of a particular asset divided by the total market cap of all cryptocurrencies. Thus, if everything goes up 5%, the market cap dominance of every asset does not change.

Looking at changes in market cap dominance gives a measure of how much the price of an asset changes, while also correcting for overall market trends.

We see that the daily and weekly Dominance changes are overall less than that of the price change. This makes sense because the last month has largely seen positive trends in the overall market.

Interestingly, we find much larger returns after a week than a day.

This is further reflected in the time it takes to reach a peak return (maximum value of market cap dominance). We find that on average, it takes about 3.5 days to achieve maximum returns after a score first crosses a score threshold. However, keep in mind that a score may stay above a certain threshold for multiple consecutive days, indicating a large buy window.

Test Strategies

Test Strategy 1: Hold any coin that has had a score above X in the last Y hours. If multiple coins should be held (multiple coins above threshold), split funds weighted by the coins’ score. 

As mentioned above high scores can 1) last multiple days and 2) take an average of 3 days to see returns. With these two ideas, it makes sense not only to consider the current scores of assets, but also recent historical scores.

This plot shows an example of how this strategy performs given a score threshold (X) of 80, and a time window (Y) of 12 hours. In other words, if a coin has had a score of 80 (or higher) in the last 12 hours, hold it. Once it has not had a score above 80 in the last 12 hours, sell. These parameters map closely to the altcoins’ overall performance for a while but results in some “homeruns” later on.

As a reasonable point of reference, we compare this strategy to the performance of holding Bitcoin (orange) and the performance of holding all alts equally (green).

We have done other comparisons, for example, randomizing scores associated with assets (i.e., shuffling all the names, so if a score is 90 on RUNE, the strategy buys ATOM instead for example).

This is a test to see if the scores are actually meaningful, or if someone can buy anything at random and perform just as well. As expected, the random shuffling approach typically performs similarly to the green line (or worse).


This shows the performance of the strategy across multiple values of score threshold (X) and time window (Y). The heat maps show the score threshold on the left axis, and time window (hours) on the bottom axis. The two metrics used here are relative final balance (left) and Sharpe* ratio (right).

The Sharpe ratio gives a measure of how consistent the strategy is, where larger values represent more consistent, positive returns. Specifically, it is the average hourly return divided by the standard deviations of hourly returns. Thus, a high Sharpe ratio can either come from very high, positive hourly returns, and/or from low standard deviations in returns.

We find that high score thresholds (90) with small look-back windows (12, 24 hours) show the largest ratios, but still relatively high final balances. This implies that they are making riskier “all-in” bets. Because 90’s come less frequently, a strategy that uses X = 90, Y = 24 will often be holding no alts (defaults to BTC when there is nothing to buy).

However, if a single alt has a score of 90, the strategy goes all in. An example of this is highlighted in the bottom plot for X = 90, Y = 24 hours. This contrasts strategies with lower scores, which more often cause funds to be distributed across all of the assets. 

*Note: Ratios like these are highly sensitive to the amount of data collected and specific assumptions made.

Test Strategy 2: Buy at score X, sell at score Y.

This strategy is quite intuitive. Buy when a score crosses a certain threshold and sell when the score crosses below another threshold. This strategy performs similarly to strategy 1 on average: However, the best performances of strategy 2 outperform strategy 1 in both balance and Sharpe.

The best returns we found between these two strategies was 3.8x, simply buying and holding a coin while its score is above 80. This performance is shown by the orange line in the bottom plot.

The highest Sharpe, however, was achieved with buy at 80, sell at 60. This strategy holds more coins at any time, on average (because it takes longer to sell). Thus, performance is not determined by single “homeruns”.

The circled region shows a good example of how the two strategies may achieve similar final performances; however the orange line shows a lot more variance. Again, this is because the orange line (buy 80, sell 80) is less diversified.

---

Cointelegraph is a publisher of financial information, not an investment adviser. We do not provide personalized or individualized investment advice. As a condition of using Markets Pro you acknowledge and agree that no Content published or otherwise provided as part of any Service constitutes a personalized recommendation or advice regarding the suitability of, or advisability of investing in, purchasing or selling any particular investment, security, portfolio, commodity, transaction or investment strategy. Cryptocurrencies are volatile investments and carry significant risk. Consult your financial advisor before making financial decisions. Full terms and conditions.