Standard deviation comes first and the mean comes last in this paper's only ranking. Neither a company nor a forecasting model gets ranked.

Where are the stock predictions?

The paper presents itself as stock price prediction for four companies. Table 1 carries the caption "Stock Price Prediction", and the commentary on subsequent tables repeats the phrase. Its premise is that better forecasts help investment decisions, "particularly those focused on short-term investments."

The proposed tool is a ranking procedure. COPRAS, named alongside AHP, is supposed to assess prediction models on multiple criteria. Yet the paper supplies no price data, sample period, frequency or exchange. The reported finding ranks five statistics.

Prasanth, Ramachandran and Sivaji frame stock forecasting as a decision problem. The abstract says they use ANNs and GANs to predict final prices. It also says they introduce a Deep Belief Network tuned by the Dragonfly Algorithm, with social media sentiment as an input (Twitter, according to the introduction). COPRAS, with AHP alongside it, would "assess and rank different prediction models based on multiple criteria". The stated payoff is better decisions for short-term investors.

COPRAS (Complex Proportional Assessment) ranks alternatives against multiple criteria. It starts by normalizing and weighting a decision matrix. For each alternative, the weighted benefit criteria are summed as Bi and the weighted cost criteria as Ci. Combining them gives a significance score Qi. Each Qi is then converted to a utility Ui, setting the best alternative to 100.

The application occupies little space. Table 1 gives five statistics (Mean, St. Error, Median, Mode, St. Deviation) for four companies: Mobil, First Bank, Nigeria Brewery and General. The paper never identifies them as listed securities. Its Materials and Method entries concern unrelated subjects: the "Mobil" entry defines the standard normal distribution, while "General" discusses the words "public" and "common". Under "First Bank" comes a history of the World Bank. "The World Bank was established to help rebuild Europe and Japan following World War II," the entry says. The authors then normalize and weight the matrix, calculate Bi, Ci, Qi and Ui, and give the final ranking in Table 8. We found no sample period, price series, frequency or data source for Table 1.

The rows get ranked

COPRAS treats rows as alternatives. In this matrix, those rows contain the five statistics. The companies occupy the columns, renamed Criterion 1 to 4 from Table 3 onward. Thus Table 8 puts St. Deviation 1, Mode 2, Median 3, St. Error 4 and Mean 5.

The accompanying text interprets the ranking as descriptive output. It reports "the mean rank is 5," and says "the standard deviation of 1 highlights the consistency in the ranking process". From there it claims that "these ranking values help in determining the most suitable stock choices".

No stock receives a rank.

Table 1 contains twenty cells, ranging from 0.12 to 0.31 in exact steps of 0.01. Mobil spans 0.12 to 0.16; First Bank, 0.17 to 0.21; Nigeria Brewery, 0.22 to 0.26; and General, 0.27 to 0.31. If St. Error means the standard error of the mean, Mobil's 0.13 and standard deviation of 0.16 imply about 1.5 observations. General's 0.28 and 0.31 imply about 1.2. The authors interpret General's 0.31 as "greater fluctuations in stock price predictions". Although Table 1 is captioned "Stock Price Prediction", none of its cells has a date, horizon or price.

Nigeria Brewery and General become costs

The authors divide each company column by its sum. Mobil totals 0.70, giving its mean a normalized value of 0.12/0.70 = 0.1714. Each column forms an evenly spaced sequence around its median. Median therefore normalizes to 0.2000 for all four companies. The paper describes this as "a balanced central tendency."

A weight of 0.65 is then applied to every cell. Across four criteria, 0.65 sums to 2.6. Uniform weights multiply every score by the same factor and leave the ranking unchanged. AHP appears in the abstract, in three literature passages and in the conclusion. We found no pairwise comparisons or derived weights in the calculation. The authors explain that "each stock is given equal importance in the analysis."

The benefit and cost assignments are unstated, though the arithmetic reveals them. For the Mean row, Bi is 0.65 × (0.1714 + 0.1789) = 0.228, using Mobil and First Bank. Ci is 0.65 × (0.1833 + 0.1862) = 0.240, using Nigeria Brewery and General. The same calculation reproduces the St. Deviation row's 0.292 and 0.280. This treats two companies as quantities to maximize and two as quantities to minimize. We found no explanation for either assignment.

The layout of the grid then determines the order. Normalization gives the lower-valued companies a wider spread: Mobil ranges from 0.1714 to 0.2286, whereas General ranges only from 0.1862 to 0.2138. Moving from Mean to St. Deviation raises Bi by 0.064 (0.228 to 0.292). Ci increases by just 0.040 (0.240 to 0.280). The bottom row wins as a result.

What utility 100 measures

St. Deviation receives Ui = 100.0000 in Table 7; Mean receives Ui = 95.3488. COPRAS assigns the 100 by construction, scaling utility relative to the best row. Across the five alternatives, Qi runs from 0.508 to 0.533, a range of 0.025. Neither measure is a forecast error or a return. A desk cannot buy "St. Deviation."

Nigerian prices would not repair the design

We did not test the paper's method. We could not reproduce it as a strategy. The paper supplies weights (0.65 everywhere) and criteria (the four company columns), yet leaves out the benefit or cost designations, an estimation window and a trading rule. The four companies appear to be Nigerian listings, outside our price-data coverage. We also lack the Twitter sentiment required by the Dragonfly-tuned network. Substituting US stocks would test a different universe.

The data gap is the smaller obstacle. Even with all the price series available, there would be nothing to backtest. We found no dated forecast or company-level ranking. The abstract claims the research "introduces a novel approach" built around the Dragonfly-optimized network. Beyond the abstract, that model appears in an introduction passage ending with a citation and in the conclusion, where it is called "proposed". No table or calculation implements it. Meanwhile, the methods text discusses coconut copra meal as a cadmium biosorbent and materials for spur gears.

"No universal solution has yet been established," the authors acknowledge. That sentence also says the research "contributes to ongoing efforts to develop more accurate forecasting models". Measuring accuracy requires a forecast and an outcome against which to score it. None of the eight tables matches a labelled prediction to a date or a realized price.

It would take little to change my view. Transpose the matrix so that the four companies become the alternatives. Identify each criterion as a benefit or a cost and explain the choice, such as expected return as a benefit and volatility as a cost. Rank the companies on a fixed date, then score the ranking against subsequent returns. Until then, Table 8 ranks row positions in a 5 × 4 grid of evenly spaced numbers.