Meta-post on usefulness of timing discussions from a statistical learning perspective

There's some confusion lately about what constitutes market "timing" and while the purpose of this thread is not to recommend any such tactics - I don't even use any myself - here I'd like to reason about why it's (probably) not a good idea for folks to engage in popular styles of speculative trading and which aspects of the definition or the underlying issues are consistent with reasoning.

1. Market timing solely based on price levels or returns tends to be severely lacking in statistical significance because if such an obvious signal could be traded on, it would easily be arbed out immediately or not reliable enough to generate alpha. In other words, such indicators have approximately zero predictive value and would otherwise be a "free lunch" in a negative-sum game.
2. Using an excessive number of parameters (e.g. combining multiple moving-average crossovers or fancy pseudo-scientific technical analysis) leads to unstable estimation, which is exacerbated by the curse of dimensionality. There's a reason why classical linear regression (or with a penalization term, as in lasso or ridge) is preferred by many quant hedge funds even by some researchers who have extensive backgrounds improving large-scale state-of-the-art ML algorithms. When you overfit so severely in-sample, the strategy can perform even worse out-of-sample than a simple regression with many degrees of freedom.
3. If you examine the market on a short time scale (depending on trading frequency and the product), the observed data will appear to have trends. In that sense, you can say that ex-post, the market seems to have structure (and this is another reason why I'm wary of Monte Carlo simulations or bootstrapped resampling of financial time series). However, even if this were true, the problem that non-stationarity poses is that there is no guarantee as to what the structure will be or how long it will last. A regime shift may occur before, during, or after an economic event somewhere else in the world when you're least expecting it - and this is especially dangerous when engineered tail risk faces black swans. You can look at charts and find an idea that would've worked great for a few months or even years (e.g. during a directional market), but then it gets crushed when a new pattern emerges contrary to your posterior beliefs.
4. Just because you can find a pattern, even if it's a long-term pattern, does not mean that you can execute on it profitably. Cost of borrowing is not a constant. The cost of trading is not a constant either. What time of the day are you trading, and how much of the profit is made around economic events? Especially during heightened volatility, spreads tend to widen and it becomes increasingly expensive to trade actively (crossing the bid-ask spread) due to low liquidity. And without level 2 data, you can't even judge the depth of the order book. The open interest at NBBO may be paper thin. After all that, in some countries such as the US, there's still the question of whether the additional profits from trading at short-term capital gains rates is worth the difference in taxation versus long-term gains.
5. I've seen many strategies discussed whose performance is summarized in a single number, e.g. CAGR or Sharpe ratio, which again has very little meaning in terms of inference. While it doesn't make much sense to compute explicit confidence or prediction intervals, I do believe there is value in examine individual months' returns. If we excluded the influential extreme events from a sample, how would the results look now? Are the recent 1 year, 5 years, 10 years, and 20 years of returns similar to those of the 1920s-2000s in an extended backtest (usually no, because markets are becoming more efficient over time).
6. Questions like "I started HFEA last year and just lost (or gained) $X on Tuesday. Should I sell security Y on Wednesday or buy more of Z?" are not genuinely helpful at all because hardly any anonymous on the Internet knows your volatility capacity, investment horizon, aggregate portfolio composition, income situation, personal expenditures, family budgeting plans, and overall lifestyle objectives. Even a person who knows where they want to be positioned going into an FOMC often can't give solid advice to another strange whom they've never met before.
7. If you're a non-systematic trader, do you have the mental fortitude to continue making the right judgment calls even the market is not behaving as you expect due to information that is not yet available to you (e.g. Erdogan's tanks suddenly rolling into the streets of Turkey at 3am in 2016, leading to a rapid collapse of the Lira)? And have you objectively measured your consistency with a sufficiently large sample against straightforward buy-and-hold? Is it worth your energy and the opportunity cost of time (or is the market merely a glorified casino for fulfilling psychological thrills)?

I realized by now that the whole post has a rather pessimistic tone, but this is not to detract from the fact that you can achieve above-average risk-adjusted returns through diversification (because of less than perfectly correlated returns that reduce portfolio volatility) and absolute returns through the use of leverage (in assets that generate a profit and have sufficiently high Sortino ratios, you're being compensated appropriately by the market for undertaking "risk"). That's the underlying premise of allocations such as HFEA and AWP in line with modern portfolio theory. In fact, it's hardly unexpected to achieve such performance metrics when you're investing closer to the efficient frontier than purely 100% equities.

Beyond such fundamentals, to "beat the market" you'll either have to find a niche opportunity that is truly not well-understood or realistically capitalizable by the majority of investors (e.g. mining early-day Bitcoin or knowledge on planned developments in local real estate through extensive personal connections). In public markets (as with the case of stocks and ETFs), there are teams of PhDs to whom all the books you can find on Amazon and articles on Arxiv only touch the tip of the iceberg, and who sharpen the edge of their systems over many years of accumulated experience in a competitive and consolidating industry while equipped with data, infrastructure, and rigorous non-public research at a level beyond the dreams of most amateur/hobbyists. And no, most people are not capable of becoming successful marketmakers or portfolio managers; the survivorship bias is enormous even at a firm level. (In an accelerating arms race into the 21st century, the vast majority of retail traders are farmers armed with axes and pitchforks whose access to information includes a few well-known verses from the Bible. Better to stay home and avoid becoming a statistical loss in the crusades.)